Crystal Shape Engineering - Industrial & Engineering Chemistry

9812

Ind. Eng. Chem. Res. 2008, 47, 9812–9833

Crystal Shape Engineering Michael A. Lovette, Andrea Robben Browning, Derek W. Griffin, Jacob P. Sizemore, Ryan C. Snyder, and Michael F. Doherty* Department of Chemical Engineering, UniVersity of California, Santa Barbara, California 93106-5080

In an industrial crystallization process, crystal shape strongly influences end-product quality and functionality, as well as downstream processing. In addition, nucleation events, solvent effects, and polymorph selection play critical roles in both the design and operation of a crystallization plant and the patentability of the product and process. Therefore, investigation of these issues, with respect to a priori prediction, is (and will continue to be) an important avenue of research. In this review, we discuss the state-of-the-art in modeling crystallization processes over a range of length scales relevant to nucleation through process design. We also identify opportunities for continued research and specific areas where significant advancements are needed. 1. Introduction Crystallization from solution is a process used in the chemical industries for the preparation of many types of solids (e.g., pharmaceutical products, chemical intermediates, specialty chemicals, catalysts). Several key properties of the resultant materials originate from this process, including chemical purity and composition, internal structure (polymorphic state), size and shape distributions, and defect density (crystallinity). Size and shape distributions impact various solid properties, including end-use efficacy (e.g., bioavailability for pharmaceuticals,1 reactivity for catalytics2), flowability, wettability,3-5 and adhesion.6 In turn, these properties impact downstream processing efficiency (e.g., filtering/drying times and the possible need for milling), as well as storage and handling. The desired size and shape distributions of a crystalline product will vary, depending on its application. For example, needle-shaped (i.e., high-aspect-ratio) crystals are well-known to be difficult to process,7 making them undesirable for pharmaceutical applications. However, needlelike shapes are key to several other applications (e.g., ZnO nanowires for solar cells8), and, in these cases, crystal shapes are engineered to be as needlelike as possible. Because of the broad impact of crystal size and shape distributions, general methodologies for the prediction, measurement, and improvement of crystal size and shape are desired for product and process design. While these topics have been studied for over one hundred years, recent breakthroughs in experimental techniques, modeling, and simulation have led to an improved understanding of crystal growth from solution. This article reviews some of these developments and presents a path for future progress. The remainder of this article is organized as follows. It begins with a review of recent advances in the understanding of nucleation phenomena, highlighting experiments and simulations that have provided insight into the size, shape, and structure of critical-sized nuclei. The shapes of larger crystals are discussed next. This discussion begins with a description of nonmechanistic models and then moves to a detailed discussion of crystal growth mechanisms. This is followed by a review of current * To whom correspondence should be addressed. Tel.: (805) 8935309. Fax: (805) 893-4731. E-mail address: [email protected].

modeling, experimental, and simulation techniques for obtaining crystal shapes. Efforts in shape modification through the use of additives are described next, followed by methods to predict dynamics of crystal shapes evolving in time. Finally, crystal shape is discussed, along with crystal size, in the context of population balance models. 2. Nucleation and Polymorph Selection The role of nucleation in polymorph selection is of particular industrial interest. The polymorph (different crystal structures of the same molecule) obtained can have a large impact on the final utility (for example, the Ritonavir case9) and patentability of a crystalline product. Different polymorphs can have different physical properties, such as stability, hardness, color, solubility, dissolution rate, and bioavailablity.10 Recent experiments have shown that the internal structure of a crystal may be set during nucleation.11 Furthermore, experiments have shown that a seeded polymorph may not appear exclusively in the final crystallization product.12 A method to predict the observed polymorph is an important goal in the study of nucleation. However, a more fundamental knowledge of nucleation must be achieved before predictive tools can be perfected. Traditionally, classical nucleation theory (CNT) has been used to describe gas, liquid, and solid nucleation.13-18 Classical nucleation theory was explained by Gibbs in the 1870s,13 and it has proven useful for understanding nucleation in a variety of systems, in particular, gases and liquids. Nonetheless, Gibbs himself questioned its validity for solid nucleation. CNT is based on the total change of the free energy of the system during a nucleation event. In a nucleation event, energy is decreased by the creation of the new lower-energy phase (see the first term in eq 1), and energy is increased via the formation of an interface between the two phases (see the second term in eq 1). For a spherical nucleus, the change in Gibbs free energy resulting from the creation of a nucleus with radius r is given by ∆G ) -

( )

4 πr3∆µ + 4πr2γ 3 VM

(1)

where VM is the molar volume of the solid nucleus, γ the specific surface energy of the solid-liquid interface (which, in this case,

10.1021/ie800900f CCC: $40.75  2008 American Chemical Society Published on Web 11/13/2008

Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 9813

Figure 1. Two-step nucleation of colloidal particles. Image (a) shows the initial liquid phase, whereas image (b) shows the dense liquid nucleus created first, followed by formation of the crystalline nucleus in the dense liquid (depicted in image (c)).31

is assumed to be isotropic (i.e., the same in all directions)), and ∆µ the difference in chemical potential (on a per mole basis) between solute in solution and in the nucleus (∆µ > 0, for supersaturated solutions). The free energy surface, ∆G(r), resulting from eq 1 has a single stationary point that occurs at a maximum in ∆G (for anisotropic nuclei, this will occur at a saddle point, as shown for two-dimensional (2D) nucleation in Figure 8, displayed later in this paper), yielding a critical-sized nucleus of radius rc, given by rc )

2γVM ∆µ

(2)

The change in ∆G of the system at this stationary point is given by ∆Gc )

( )

16 VM 2 3 4 2 π γ ) πrc γ 3 ∆µ 3

(3)

If a cluster is larger than this critical size, it has a higher probability to grow than decay; if the cluster is smaller than this critical size, it has a higher probability to decay. This indicates that nucleation (as predicted by CNT) is a thermally activated process with a free-energy barrier height of ∆Gc, which, therefore, occurs at a rate given by

( )

I ) κ3D exp -

∆Gc kBT

(4)

where I is the nucleation rate (which has dimensions of the number of nuclei per unit volume per unit time), κ3D a kinetic prefactor, kB the Boltzmann constant, and T the temperature.19 As shown in eq 2, the critical nucleus size is affected by surface energy, supersaturation, and nucleus shape. As the supersaturation increases and the system moves away from equilibrium (∆µv), the critical nucleus size will decrease. It is important to note that the shape and polymorph of the nucleus (the latter of which impacts γ) must be specified a priori, in using CNT to determine rc and ∆Gc. It cannot be assumed that the surface energy of a critical nucleus is identical to that of the bulk solid. This is due to the microscopic size of the nuclei (O) (1 nm), and possible differences between their crystal structures and those in the macroscopic solids.20 This has been supported by recent simulations of NaCl21 and SF6 crystallization.22 However, the magnitude of the deviation in surface energy between criticalsized nuclei and bulk solids is not known. In addition to the differences in overall structure, the nucleus may have a different

shape than the final crystal, revealing different faces and their different surface energies. Furthermore, the nucleus may even be spherical/elliptical in shape, in which case the concept of faces does not apply. Because critical nuclei are small, no direct experimental surface energy measurements have been made on them. This causes all theories and ideas about the correct nucleus surface energy to remain experimentally unverified. In addition to the question of surface energy, surface work terms, such as surface stress, may also be required to be included for an accurate theory of solid nucleation.23 Surface stress, although unimportant in fluid nucleation, may become significant when the nucleus surface can experience strain. In recent years, the theory of two-step crystallization (which is also called two-step nucleation) has attracted attention. In two-step crystallization, nucleation proceeds through a dense liquid (amorphous) step before ordering into the growth structure.24 This theory is supported by both simulations25-28 and experiments.29-31 Two-step crystallization has been experimentally observed in protein and colloidal systems and in simulations of MnO2. The two-step progression from liquid to crystalline nuclei observed in colloid experiments can be observed in Figure 1. Because CNT assumes a single step, the evidence supporting two-step crystallization raises the question of how to include a multistep process in nucleation theory.29 Experimental studies and simulations are being actively pursued to enable methods of nucleus polymorph and shape prediction. Critical nucleus size can be approximated in simulations and experiments by finding the size where the nucleus begins to have a higher probability to grow than decay. For example, in the experiments of Yau and Vekilov,11 they determined that clusters were just above the critical size if the detachment rate was significantly smaller than the attachment rate and clusters were just below the critical size if the attachment rate smaller than the detachment rate. In simulations such as those performed by Frenkel and co-workers,32,33 the critical size was determined by finding the maximum in Gibbs free-energy difference of nucleation, as a function of nucleus size, as done in CNT. However, nuclei are difficult to observe experimentally as they are often indistinguishable from their parent solution due to their small size (O (100 particles)) and low concentration. Frenkel and co-workers34 estimated that, for melt NaCl crystallization, a single nucleus would form in a droplet of diameter O (3 µm) in O (10 s) which highlights the rarity of nucleation in terms of experimental observation.

9814 Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008

Figure 2. (a) Atomic force microscopy (AFM) image of a critical nucleus of apoferritin molecules with molecules in nucleus shown in green (image adapted from Yau and Vekilov11). (b) Depiction of a critical nucleus of poly(methyl methacrylate) (PMMA) colloids (with particles in the nucleus shown in red and particles with only one bond to the nucleus shown in blue; the size has been reduced, for the sake of clarity) (image adapted from Gasser et al.35). Reprinted with permission from AAAS.

Designing experiments to ensure that nuclei are observed before significant growth has occurred remains a challenge. These challenges have been overcome in large particle experiments, where nucleation has been directly observed.11,35 Colloids and globular proteins are large enough and nucleate slowly enough for nucleation to be observed using techniques such as optical and atomic force microscopy (AFM). The colloidal and protein nucleation experiments performed by Gasser et al.35 and Yau and Vekilov,11 respectively, found critical nuclei with 25-100 particles and nonspherical shapes, as shown in Figure 2. In both cases the internal structure of the critical nuclei corresponded to the thermodynamically stable polymorphs.10 Small-angle neutron scattering (SANS) techniques have also been used to observe nucleation of polymer blends.36 The results of these experiments have provided new insights into the nucleation process. For small molecules, direct measurement and observation of nuclei is currently impossible. Although limited by the small length and short time scales, indirect measurements of nucleation can be made.37-40 Indirect methods include the observation of crystal particles after growth to a larger size and characterization of the final crystal structure, with attempts made to keep the nucleation rate low after initial nuclei formation. These methods can measure average nucleation rates, induction times, and the effects of experimental conditions on crystal structures. The development of new computational techniques coupled with increases in computing power have recently allowed for nucleation phenomena to be simulated. To date, most nucleation simulations have been performed using model particles with simple pair-additive potentials.41-47 Some nucleation simulations have also been performed using many-body potentials.48-50 The model particle studies performed by Browning et al.47 (in which the solvent was modeled implicitly by charge screening) found critical nuclei similar to the protein and colloid experiments, being composed of ∼100 particles and having a nonspherical shape, as shown in Figure 3. In these simulations, the majority of the critical nucleus’s structure corresponded to that of the stable polymorph; however, some percentage of the metastable polymorph was also found. While this example showed that polymorph selection takes place during nucleation, the interaction potential between particles51 and the thermodynamic conditions of crystallization52 can effect when selection occurs.

Figure 3. Near-critical-size nuclei found in model particle simulations performed by Browning et al.47 Copyright 2008. American Physical Society.

Nucleation simulations using interaction potentials based on small molecules have also been performed. The critical nucleus of ice in supercooled water was determined to consist of 210-260 molecules.53 Simulations of amorphous silicon found critical nuclei that consisted of 30-50 atoms (∼1.2 nm in diameter) with an almost spherical shape.54 Critical-sized nuclei of N2 were found to contain ∼300 molecules.55 Faceted critical nuclei of similar size have been observed in melt nucleation simulations of NaCl34 and SF6.22 A summary of selected critical nucleus sizes and shapes is reported in Table 1. Because of the time and size scale difficulties that are involved with simulating nucleation, precritical aggregates have also been studied. Simulations of NaCl in water showed that the centers of stable aggregates had nonhydrated Na+ ions.56 Acetic acid aggregation in carbon tetrachloride (CCl4) solution was observed to proceed through dimer formation to oligomers to hydrogen-bonded micelles.57 The solvents were modeled explicitly in both the NaCl and acetic acid simulations. New simulation techniques such as reaction coordinate calculation and transition path sampling58,59 may soon allow for a more complete study of small molecule nucleation from solution. The nucleation stage, wherein pre- and post-critical-sized clusters (containing ∼(100-10000) molecules and being ∼10 nm in diameter) grow by aggregation, is now being actively explored using molecular dynamics (MD), Monte Carlo (MC), and other simulation techniques. However, after a crystallite has grown to a microscopic size (on the order of millions of particles and ∼1000 nm in size), other methods of growth (e.g., 2D nucleation and spiral growth) have been observed. These methods are discussed in the remainder of this article. What occurs during the transition between the nucleation and mac-

Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 9815 Table 1. Summary of Selected Critical Nucleus Sizes Determined by Experiment (Apoferritin) and Simulations (SF6, NaCl, and Lennard-Jones Particles), Showing the General Agreement between the Different Systems and the Decrease in Critical Nucleus Size with Increasing Driving Force driving forcea 43.7 3.95 2.00 0.20 0.13 0.086 0.25 0.22 0.2 0.25

critical nucleus size (number of particles)

shape

Apoferritin (ref 11) 4 25 60 SF6 (ref 22) 40 120 266 NaCl (ref 34) 120 150 Lennard-Jones (ref 45) 240 Ice (ref 53) 210-260

faceted faceted faceted faceted faceted cubic (faceted) cubic (faceted) ellipsoidal

Driving force is defined as (C/Ceq) - 1 for apoferritin and 1 (T/Tmelt) for the remaining systems. a

roscopic growth stages is presently unknown. The methods that have been used to study nucleation are not sufficient to study this regime; therefore, new methods must be developed, which will surely result in a lively and dynamic research area in the next decade. 3. Equilibrium and Growth Shapes Equilibrium criteria for the interfaces that divide the solid and fluid phases were developed by Gibbs.13 Wulff developed a geometric approach for determining the shapes of faceted crystals (at constant temperature and pressure) with anisotropic surface free energies conforming to these criteria, known as the Wulff construction.60,61 The work of formation of a solid polyhedron with volume V from a fluid phase is given by the Gibbs-Thomson formula: V∆µ + ∆G ) VM

∑γA

i i

(5)

i

where γi and Ai are the respective surface free energy and area of facet i.62 The equilibrium shape occurs when ∑iγiAi is a minimum for a fixed crystal volume V and can be determined from

∑ γ dA ) 0 i

i

(6)

i

The Wulff construction (which yields the equilibrium shape) is obtained by connecting the exterior end points of vectors, with origins located at the center of the crystal, to planes with angles perpendicular to them and magnitudes Hi, proportional to their corresponding surface free energies, resulting in62,63 γi γ1 γ2 ) ) ··· ) H1 H2 Hi

(7)

Although Gibbs was the first to develop the criteria for equilibrium crystal shapes, he recognized, as footnoted in his collected works (page 325 of ref 13), that crystal shapes are usually determined by kinetics rather than by thermodynamics alone: “On the whole it seems not improbable that the form of very minute crystals in equilibrium with solvents is principally

determined [...] by the condition that ΣiγiAi shall be a minimum for the volume of the crystal, [...] but as they grow larger (in a solvent no more supersaturated than is necessary to make them grow at all), the deposition of new matter on the different surfaces will be determined more by the nature (orientation) of the surfaces and less by their size and relations to the surrounding surfaces. As a final result, a large crystal, thus formed, will generally be bounded by those surfaces alone on which the deposit of new matter takes place least readily (i.e., the slow-growing faces) [...] But the relative development of the different kinds of sides [...] will not be such as to make ΣiγiAi a minimum.” In the previous quotation, the words shown in italic are ours and the symbol “[...]” represent words that we omitted, for the sake of brevity and/or clarity. In contrast to fluids, the resistance to deformations that are ubiquitous with solids requires that shape changes occur through the deposition and removal of material at crystal faces. Therefore, the shapes of growing crystals will be determined by the net rate of material incorporation into individual surfaces rather than by their equilibrium shapes. If the net rate of material incorporation is constant on each face, a crystal will evolve toward its steady state growth shape, at which point growth proceeds in a self-similar manner. This steady state growth shape is generally not equivalent to the equilibrium shape. Analogous to the Wulff construction for predicting equilibrium shapes, the Frank-Chernov condition64,65 can be used to predict steady-state growth shapes for faceted crystals, provided the crystallographic data and growth rates of the faces are known. Thus, the remaining sections of this article (with the exception of section 8, “Coupling Size and Shape Predictions”) will focus on the development of theories, simulations, and experiments that have contributed to the elucidation of crystal growth kinetics. The Frank-Chernov condition has the same form as the Wulff construction, with the exception that the shape is determined from the perpendicular growth rates of faces, rather than surface energies: Gi G1 G2 ) ) ··· ) H1 H2 Hi

(8)

where Gi is the perpendicular growth of face i. In addition, methods to predict shape evolution in both growth66,67 (discussed in section 7) and dissolution67 have been developed. The first approaches for predicting growth shapes were based solely on the structure and interactions within the crystal. Bravais68 proposed a quantitative relationship for predicting crystal shape, based on crystal structure, which gained validity from the observations of Friedel.69 The Bravais relationship is given as Ghkl ∝

1 dhkl

(9)

where Ghkl and dhkl are the perpendicular growth rate and interplanar spacing, respectively, of the face specified by the index hkl. Donnay and Harker70 modified this approach by including reductions in dhkl that were due to extinction conditions. These conditions are the same as those that cause extinction in X-ray diffraction (XRD) patterns. This model (which is referenced as the BFDH model) represents the most easily implemented method for shape prediction, because it requires only a knowledge of the crystallography of the crystal. Later, the attachment energy model, developed by Hartman and Perdock,71,72 took into account the energetics of crystal interactions in addition to the crystal geometry. They assumed


that the time needed for the formation of a bond decreases as the bond energy increases. Defining the attachment energy, Eatt hkl, as “the bond energy released when one building unit is attached to the surface of a crystal face,” this assumption results in the perpendicular growth rates of a crystal face increasing as the attachment energy increases, yielding the relationship Ematt > Eatt n w Gm > Gn

(10)

where m and n are indices that describe two different faces. For a crystal face, the attachment energy can be determined from att Elatt ) Eslice hkl + Ehkl

(11)

where Elatt is the total energy of intermolecular interactions per slice mole in a crystal and Ehkl is the energy of the intermolecular interactions contained within a layer of thickness dhkl.73 To obtain the perpendicular growth rates from eq 10, an assumption of linearity is often made,74 i.e., Ghkl ∝ Eatt hkl

(12)

Since its inception, the attachment energy model has been further modified75 and is used extensively to predict crystal shape. Recently, however, there has been a growing trend away from the BFDH and attachment energy models toward higherfidelity mechanistic models. These models seem to be more accurate and versatile. 4. Growth Processes Although the BFDH and attachment energy models can provide accurate predictions for vapor-grown crystals, they account only for the influence of solid-side energetics on crystal shape. In doing so, they ignore the influence of external factors on crystal growth, such as solvent, additives, and supersaturation, which can impact their accuracy to varying degrees for solution crystallization. An understanding of growth processes, namely, the creation of kink sites and transport of material to growing faces, is required to develop models that can account for these factors. Kink sites are the positions on a crystal where molecules can most readily incorporate into or detach from the lattice and appear as exposed trihedral corners (shown as gray cubes in Figure 4b). At a classical kink site, the adjacent molecule in the crystal is bound by half of the possible intermolecular interactions for a molecule incorporated into the lattice,76,77 thus their description as “half crystal positions”. The method by which kink sites are created on a crystal face is dependent on solid-state factors and the state of the growth environment. During the growth of a crystal face from solution, the following processes occur in series: (1) Solute molecules are transported from the bulk solution toward the face by convection and diffusion; (2) Solute molecules and kink sites shed their surrounding solvent molecules (desolvate) and solute molecules are incorporated into kink sites; and (3) The latent heat of crystallization is released and transported to the crystal and solution. For the sake of clarity, process 1 is hereafter referred to as “bulk transport”, process 2 is referred to as “surface integration”, and process 3 is referred to as “heat transport”. The growth rate of a crystal face is limited by the slowest of these processes.78,79 Kink sites are created at a certain density on the face by processes occurring simultaneously with processes

Figure 4. Schematics of step edges at (a) 0 K and (b and c) above 0 K. The gray squares in schematic (b) represent kink sites separated by an average distance of x0. Image (c) is a schematic of layered growth of the {hkl} face growing at a perpendicular growth rate, Ghkl, through the lateral spreading of steps separated by an interstep distance y, with a height h, at a step velocity of V.

(1)-(3). The density of kink sites is included as a contribution to the rate of surface integration. One important description of the solid-state conditions for a molecular crystal is its (noncovalent) bond structure, which can be discussed using periodic bond chain (PBC) theory, as proposed by Hartman and Perdok.71 PBCs have been defined as a set of connections between growth units at their centers of mass following only “strong” interactions with an overall periodicity [uVw] ) ua + Vb+ wc, (u, V, w ∈ Z), where a, b, and c are the primitive lattice vectors coincident with the unitcell directions and Z consists of all positive and negative integers.74 Using PBC theory, faces can be classified based on the number of coplanar PBCs they contain within a slice of thickness dhkl. If a face contains at least two PBCs, it is stable and macroscopically flat (F face); if it has only one PBC then it is stepped (S face) where each step can be readily kinked; if it has none, then it is completely kinked and rough (K face). These faces are shown in Figure 5. The mechanism by which a face grows can also be dependent on the state of the external environment, specifically the supersaturation and temperature at the surface. Generally, a face will grow by whichever process (or combination of processes) provides the fastest growth rate for a given state of the growth environment. While S and K faces grow via rough growth (described in section 4.1) at all states of the growth environment, F faces grow via a layered growth mechanism (described in section 4.2) until the state of the growth environment is above a roughening transition (described in section 4.1). This is shown schematically for F faces in Figure 6, which assumes that the growth rate is limited by the rate of surface integration for all growth mechanisms. The transition between growth mechanisms has been illustrated experimentally using in situ atomic force microscopy (AFM) by Land and De Yoreo.80 4.1. Rough Growth. Some faces, such as the {111} faces of a Kossel crystal (the classic example of a K face, shown in Figure 5) are composed almost entirely of kink sites. On these faces, molecules can incorporate into the crystal, regardless of their position, and growth is typically limited by the transport


Figure 6. Growth mechanisms for a flat (F) face, as a function of supersaturation. The solid line is the growth rate. The short dashed lines are the growth rates if two-dimensional (2D) nucleation continues to be dominant below its applicable supersaturation range. The long dashed line is the rate if spiral growth was the persistent mechanism above its applicable supersaturation range.

Figure 5. F, S, and K faces for a crystal with three periodic bond chains (PBCs) (directions indicated in red).

of solute molecules to the surface. The resulting growth rate, as derived in the Supporting Information, is given by GBT ≈ VM(C∞ - CS)

D δd

(13)

where GBT is the growth rate that is limited by the bulk transport of material to the surface, C∞ the bulk concentration of the solute in solution, CS the solute concentration at the surface, D the diffusion coefficient of the solute in the solvent, and δd is the thickness of the depletion boundary layer. When growth is limited solely by bulk transport, the solute concentration at the surface is maintained near the equilibrium saturation concentration, Ceq, and eq 13 can be expressed as GBT ≈ VM(C∞ - Ceq)

D δd

(14)

Typical values for molecular organic crystals grown from solution in an industrial crystallizer are VM(C∞ - Ceq) ≈ 0.005, D ≈ 10-5 cm2/s and δd ≈ O (1 µm), result in GBT ≈ O (5 µm/s), which is ∼10-100 times larger than typical growth rates for faces that are growing via a layered growth mechanism. As a result of this large growth rate, rough faces are typically shortlived in the evolution of a crystal’s shape as they grow into edges or vertices.66 Needle-shaped crystals are an industrially important exception, as rough faces are present on (and are responsible for) the highly elongated steady-state shape of needles. F faces will also grow by rough growth at supersaturations or temperatures beyond their kinetic and thermodynamic roughening transitions, respectively. These roughening transitions have been extensively studied and the reader is referred to Chernov81 for further discussion. 4.2. Layered Growth. For F faces below their roughening transitions, kink sites are found along edges that result from either 2D nuclei forming or dislocations emerging on the surface. Because growth units are incorporated into these kink sites, the edges flow outward from their sources, forming complete layers of the crystal face. In a supersaturated solution, this process repeats, resulting in the perpendicular growth of the face. Layered growth is a slower process than rough growth, because the density of kink sites is much lower on an F face than on a rough face.

According to a theory that was developed by Frenkel82 and extended by Burton et al.,83 at 0 K, steps will be as straight as geometrically possible; however, at any higher temperatures, they will contain kink sites (Figures 4a and b). If the kink sites along an edge are formed by thermal fluctuations (in contrast to onedimensional nucleation; for more information on this subject, the reader is referred to the work of Zhang and Nancollas84 and Cuppen et al.85,86) and they are located in their most probable configuration; they will be Boltzmann-distributed along the steps with the distance between kinks (x0) being equal to

[

x0 ) ae 1 +

( )]

φkink 1 exp 2 kBT

(15)

where ae is the lattice spacing in the direction along the edge (see Figure 4b) and φkink is the (positive) energy required to form a kink.81 The energy to form a kink includes both solid-solid and solid-solvent interactions. Layered growth occurs either via 2D nucleation and growth or via spiral growth, depending on the method of edge generation. In 2D nucleation and growth, 2D nuclei that form at the surface provide edges; however, in spiral growth, screw dislocations at the surface are constant sources of edges. 4.2.1. Two-Dimensional Nucleation and Growth. Crystal faces grow via the formation of critical-sized 2D nuclei in the absence of surface-terminating screw dislocations or at supersaturations above which growth by 2D nuclei becomes faster than spirals (a method for determining this point is described in section 5.1.1). These nuclei form when individual growth units (e.g., atoms, molecules, dimers) adsorb onto the crystal face, diffuse, and agglomerate. After a nucleus has grown larger than its critical size, it becomes thermodynamically favorable for growth units to attach themselves to this nucleus. In a supersaturated solution, a larger-than-critical-sized 2D nucleus will spread laterally across the face until it reaches a boundary, which can be the edge of the crystal, the front of the layer beneath it, or a growth front from another nucleus (or spiral). 4.2.2. Spiral Growth. In the presence of screw dislocations, crystal faces grow via the spiral growth mechanism at supersaturations lower than the value where 2D nucleation and growth is the dominant mechanism. In spiral growth, screw dislocations that terminate on a crystal face act as continuous sources of edges (which will also be called steps in the discussion of this mechanism). At each dislocation, an edge is exposed to the solution. If its length is greater than a critical value, lc, and the solution is supersaturated at the surface, there will be a net incorporation of growth units into kink sites distributed along the edge. This incorporation causes the migration of the initial


Figure 7. Schematics of the first revolution of a square spiral ((a) top view and (b) perspective view). The steps on this spiral obey the following conditions: (i) Vi ) 0 for l < lc and Vi ) V∞,i for l > lc; (ii) Vi ) Vj; and (iii) lc,i ) lc,j, resulting in interstep distances yi ) 4lc.

step in its outward normal direction. As the step moves, it exposes another edge to the solution. This new step is pinned at the point where the dislocation emerged on the surface. The angle between steps is determined by the bond structure of the face. When this newly exposed step has reached its critical length, it grows in its outward normal direction and the process repeats, forming a spiral. The number of sides contained in the spiral is determined by the bond structure of the face and the velocities and critical lengths of the spiral sides. In a supersaturated solution the steps will continue to flow outward until they encounter a boundary, which can be a step from another spiral or 2D nucleus, the step located a layer below it, or the edge of the face; moreover, new steps will continually be generated at the dislocation. The process of spiral growth is illustrated in Figure 7. 5. Obtaining Shapes from Mechanistic Models, Simulations, and Experiments Crystal shape engineering, which can be defined as the ability to design processes through the modification of the crystallization environment to obtain desired crystal shapes, has benefited from the development of mechanistic models, computer simulations, and experiments investigating crystal growth processes. These approaches, which have been used to obtain/predict crystal growth kinetics, have relied on each other in their development and will continue to do so. 5.1. Mechanistic Crystal Growth Models. Mechanistic growth models predict growth rates for crystal faces by accounting for surface integration kinetics. These models are useful for regimes where growth is limited by (i) surface integration alone, or by (ii) both bulk transport and surface integration. For most molecular organic crystals grown from

solution in an industrial crystallizer, this regime encompasses the states of the growth environment for which F faces grow as spirals, as will be shown in section 5.1.2. However, when growth is limited by bulk transport, the growth rate of the face is given by eq 13, regardless of the growth mechanism. Mechanistic growth models are attractive for future development, because they provide the ability to account for the influence of surface structure and external conditions (supersaturation, solvent, additives, etc.) on crystal growth rates in a physically realistic manner. 5.1.1. Modeling Two-Dimensional Nucleation and Growth. Two-dimensional nucleation and growth models were the first mechanistic models used to describe crystal growth.76,77,87 The principle thesis of these models is that otherwise perfectly flat crystal faces grow via the formation of critical-sized 2D nuclei and their subsequent lateral growth. The growth rate of a crystal face growing by this mechanism is dependent on the size of nuclei formed and their rates of formation and lateral spreading. Classical nucleation theory has been used elsewhere to derive the change in free energy that is associated with the creation of critical-sized 2D nuclei of different shapes.81,83,88,89 A criticalsized 2D nucleus is an entity that exists in an unstable equilibrium with the system, occurring at a maximum or saddle point on the free-energy landscape. As with three-dimensional (3D) nucleation, the growth of subcritical-sized 2D nuclei occurs at an energetic expense while decay is an energetically favorable process. Conversely, the growth of larger than critically sized 2D nuclei is energetically favorable, whereas their decay occurs at an energetic expense. When the surface energy is isotropic on a face, disk-shaped nuclei are expected to form. The expressions for ∆G, rc, and


( )

I ) κ2D exp -

∆Gc kBT

(20)

where κ2D is a kinetic prefactor. Ohara and Reid88 determined an expression for κ2D, obtaining I)

( ) ( )

2Vjn12 VMσ π NAh

1⁄2

exp -

∆Gc kBT

(21)

where Vj is the speed of a molecule striking a nucleus, n1 is the concentration of unbound growth units on the surface, σ ≡ ∆µ/ (RT), and NA is Avogadro’s number. The ratio of the rates of lateral spreading to the formation of critical-sized nuclei determines the specific 2D nucleation and growth model that can be applied to characterize the growth of a crystal face. At relatively slow, although non-negligible, rates of lateral spreading, the birth and spread model can be applied. The assumptions used in this model are (i) no intergrowth between nuclei, (ii) the lateral spreading rate is constant and independent of direction, and (iii) nuclei can be formed anywhere (i.e., on top of complete or incomplete layers). Ohara and Reid88 determined the resulting growth rate for this model to be G ) hV2 ⁄ 3I1 ⁄ 3 Figure 8. (a) Geometry and (b) Gibbs free energy surface for an anisotropic parallelogram-shaped 2D nucleus. The lines are contours or equal to ∆G, and the dot is the position of a critical-sized 2D nucleus, which occurs at a saddle point in the surface.

∆Gc are similar to eqs 1-3, except appropriately modified for two dimensions.89 Interfacial energies along different directions on the faces of most molecular crystals are anisotropic, resulting in the formation of noncircular 2D nuclei. A parallelogramshaped 2D nucleus with anisotropic interfacial energies, γedge,1 ) γedge,3 * edge,2 ) γedge,4, is shown in Figure 8a. For this case, ∆G is given by

[

∆G ) h 2(l1γedge,1 + l2γedge,2) -

(l1l2) sin(ξ) VM

]

∆µ

(16)

where ξ is the angle between sides 1 and 2, and l1 and l2 are the lengths of sides 1 and 2, respectively. In this case, the critical-sized nucleus occurs at a saddle point in the free-energy landscape, as shown in Figure 8b, and the critical lengths are given by l1,c )

2VMγedge,2 sin(ξ)∆µ

(17)

and 2VMγedge,1 (18) sin(ξ)∆µ The shape of the critical nucleus is equal to the equilibrium shape, because it agrees with the Wulff construction (i.e., l1,cγedge,1 ) l2,cγedge,2 or γedge,1/H1 ) γedge,2/H2, where H1 ) (l2,c/2) sin(ξ) and H2 ) (l1,c/2) sin(ξ)). The resulting free energy of formation of the critical-sized nucleus is given by l2,c )

(

)

4VMh ∆Gc ) γ γ (19) sin(ξ)∆µ edge,1 edge,2 As with 3D nucleation (treated in section 2), 2D nucleation is a thermally activated process with a free-energy barrier height of ∆Gc. Thus, the nucleation rates (denoted as I and given in units of nuclei per unit area per unit time) will be given by

(22)

where V is the rate of lateral spreading (also called the step velocity); methods for determining V are given in section 5.1.3. Cuppen et al.90 has provided a more-detailed analysis of the birth and spread model for rectangular-shaped nuclei with directionally dependent rates of lateral spreading. To obtain realistic growth rates for faces growing via this mechanism (G J 0.1 nm/s), eq 21 limits the range of ∆Gc to small orders of magnitude (which implies large ∆µ or small γedge/(RT)), because larger values of ∆Gc result in insignificant nucleation rates. Land and De Yoreo80 reported the lowest values of σ at which growth by 2D nucleation was observed experimentally for various compounds, which are reproduced (in terms of R, where R ≡ C/Ceq) in Table 2. The apparent correlation between normalized edge energy and the onset of 2D nucleation and growth as the dominant growth mechanism is shown in Figure 9. When small organic molecules are crystallized by cooling, the upper limit on supersaturation is normally C/Ceq j 1.25, which is shown in the gray region in Figure 9. Therefore, most F faces on organic crystals grown from a cooled environment are expected to grow via the spiral mechanism. 5.1.2. Modeling Spiral Growth. In 1931, Volmer and Schultze92 observed the growth of iodine crystals at R ≈ 1.01, which is far below the minimum supersaturation predicted for growth to occur by the 2D nucleation and growth mechanism. This study demonstrated the need for an alternative mechanism to explain layered growth at low supersaturations. This alternative came in the form of the spiral growth mechanism proposed Table 2. Edge Energies and Minimum Values of r for Observed Growth by 2D Nucleation for Various Compoundsa atoms per γedge molecule molecule h (nm) (erg/cm2) CaCO3 KDP (101) thaumatin canavalin catalase

5 8 1552 10660 16436

0.31 0.51 5.5 7.4 11.7

675 20 0.4 2 0.32

γedgeahb (erg)

R value required for 2D nucleation

Ncc

1.3 × 10-12 4.2 × 10-14 1.3 × 10-13 1.2 × 10-12 5.2 × 10-13

20.1 1.1 1.2 12.2 2.5

lc )

(32)

The velocity profiles (eqs 30 and 32) approach each other as the supersaturation increases. The effects of different velocity profiles on growth rates were first discussed by Burton et al.83 Recently, Chernov104,105 has examined the validity of eqs 31 and 32 in light of in situ AFM observations of growing steps on various materials.103,106,107 Observations of step velocities for calcite performed by Teng et al.103 are reproduced here in Figure 10a. As shown in the figure, the velocity profile is independent of supersaturation, which is not consistent with the velocity profile predicted using the Gibbs-Thomson relationship (eq 30). Interstep distances can be determined from the geometries (including angles and critical lengths) and velocities (including velocity profiles) of a spiral’s steps.94 In doing so, Burton et al.83 assumed a round Archimedean spiral comprised of steps with velocity profiles consistent with eq 31. Applying these assumptions, they found y ) 4πrc for their calculation of

interstep distance, where rc is the critical radius of a disk-shaped 2D nucleus. Cabrera and Levine108 determined y ) 19rc for similar spirals; this relationship was used by Liu et al.97 for crystal shape predictions and is commonly cited throughout the literature. Because rc ≈ σ-1, y ≈ σ-1 using either the Burton et al.83 prediction or the Cabrera and Levine108 prediction. Teng et al.103 and Chernov105 have discussed the effect of the velocity profile on isotropic square spirals. For square spirals with step velocity profiles that conform to Voronkov’s profile (eq 32), y ) 4lc (see Figure 7), whereas square spirals with step velocity profiles that conform to eq 31 result in y = 9.5lc. In situ AFM observations of calcite growth103 have demonstrated steps advancing with y ≈ σ-1 (Figure 10b) and y ≈ 4lc, indicating the usefulness of Voronkov’s102 relationship. Assuming square spirals that obey eq 32, interstep distances will be within the range of y ) O (100-1000) nm for γedge ) 25-100 erg/cm2, VM ) 150 cm3/mol, and σ ) 0.01-0.1. 5.2. Crystal Simulations. Simulation techniques are also valuable tools for use in crystal growth modeling. Some of the earliest simulation results for crystal growth were those of Gilmer and Bennema, where the dynamics and transition probabilities of surface diffusion in crystal growth were determined for the (100) face of a Kossel crystal.109 Since that time, simulation techniquessincluding Monte Carlo (MC), kinetic Monte Carlo (KMC), and molecular dynamics (MD) simulationsshave been applied to crystal growth for a variety of molecules. (See the reviews of Rohl110 and van der Eerden111 for the development of simulation techniques.) More recently, several groups have continued to advance the field of crystal growth via simulations. The Meekes group has continued to evolve their MC algorithm to include more-complex and generic systems such as fats.112 Gilmer continues to develop crystal growth simulations using a lattice MC approach, most recently for a high explosive pentaerythritol tetranitrate (PETN).113 In addition, Piana et al. have developed a KMC scheme and they have applied it to the growth of urea crystals from water.114 A similar scheme was followed by Anderson et al.115 to predict the shape of hydrothermally synthesized zeolite A crystals. These simulation techniques have both advantages and drawbacks, compared to the modeling methods described in section

9822 Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 Table 4. Microscopy Techniques Used To Determine Crystallization Kinetics and Growth Mechanisms technique

information obtained

optical microscopy

• applied for in situ or ex situ characterizations of morphology • characterize additive/solvent interactions with specific faces • identify growth mechanisms (layered or rough) • measure crystal face growth rates • monitor shape evolution and polymorph transformations

interferometry

• applied for in situ examination of morphology and surface steepness • simultaneously measure step velocities and face growth rates • identify growth mechanisms (layered or rough) • determine solute concentration iso-contours and depletion boundary layer thicknesses

electron microscopy

• applied for ex situ examination of morphology and surface topography • characterize shape and additive/solvent interactions with specific faces • identify growth mechanisms (spirals, 2D nuclei or rough)

atomic force microscopy

• applied for in situ or ex situ characterizations of surface topography • characterize additive/solvent interactions with specific faces • identify growth mechanisms (spirals, 2D nuclei or rough) • determine step kinetic coefficients • determine kink densities (macromolecular systems)

5.1 for problems ranging from fundamental discovery to product and process development. MD simulations can be extremely useful in determining the fundamental mechanisms and movements of molecules as they are incorporated into the crystal. The difficulty in applying these MD schemes to crystal growth is the disparity in the relevant time and length scales between processes. For example, considering a molecular crystal surface of O (1000) molecules O (10 by 10) nm, a MD simulation of 1 ns would require approximately O (10) hours of single processor computation time. KMC schemes can allow for calculations on more-realistic time and length scales (micrometer-sized crystals in O (10) hours of computation time); however, to implement these schemes, the MC moves must be defined. These moves are often determined using underlying MD schemes113,114 or other movedetermining methods that require previous knowledge or good intuition about the specific system. Thus, for calculations to be performed over a range of conditions (i.e., multiple solvents), the MD and KMC codes must be rerun. It is this set of challenges of time scales that limits the ability to interlink simulation methods into process and product design methodologies. However, simulation techniques are excellent tools for discovering the underlying physics and solid-state chemistry of the processes that occur during crystal growth. The insight gained can be used to improve the fidelity of mechanistic models described in section 5.1. 5.3. Experiments. A variety of microscopy techniques have been used to investigate crystallization phenomena. A summary of these techniques, adapted from ref 116, is given in Table 4. Optical microscopy is a commonly used technique to characterize the morphology of crystals J1 µm in size, both in

situ and ex situ. It is often the first microscopy technique used to characterize a crystallization product. Many temperaturecontrolled flow-cell investigations have been performed in situ to determine the growth shapes of various small molecule molecular organic crystals.117-119 The kinetics of polymorphic phase transformations in drug solutions have been studied in situ by Veesler et al.,120 using a quiescent thermostatted batch crystallizer placed under an optical microscope. Although the spatial resolution of optical microscopy is limited by the wavelength of light, it offers the ability to image growth processes in situ at the length scale equivalent to that of crystal faces. Interferometry provides the ability to measure the steepness of the surface and face growth (or dissolution) rates simultaneously. Interferometry provides a higher vertical resolution (∼1 nm)121 than optical microscopy, allows for faster growth rates, and can be used to study larger areas than in situ AFM.122 However, the technique is limited by relatively poor lateral resolution (∼1 µm), making it unable to resolve individual steps.122,123 Interferometric studies have been performed on several macromolecular crystals,101,124 KH2PO4 (ADP),125,126 and small-molecule organic crystals (such as paracetamol,99 among others). Unique to interferometry is the ability to observe solute concentration iso-contours near crystal surfaces. These observations allow for the determination of depletion boundary layer thicknesses.126 Investigations into solution crystal growth are limited to ex situ observations when using any of the various electron microscopy techniques, because they are all operated under vacuum. Electron microscopy (specifically scanning electron microscopy (SEM)) has traditionally been used to determine crystal shapes. However, modern field-emission scanning electron microscopy (FE-SEM) systems can use low accelerating voltages (∼0.5-1.5 kV), which enable the study of surface topography on nonconducting materials. Low-acceleratingvoltage FE-SEM studies have provided clear images of growth terraces and spirals on faces of zeolite and zeo-type materials.127-129 These studies have been important in determining growth mechanisms for these systems.128-130 AFM has been used to characterize the surface microstructure of a variety of different crystalline systems, including ionic crystals,103,122,131-137 porous materials,115,128,130,138-141 macromoleculesandproteins,142-144 andsmall-moleculeorganics.145-148 AFM provides the ability to image crystal growth and dissolution processes in situ at nanometer-scale resolutions. This ability has been used to investigate several fundamental aspects of crystal growth, including methods of kink formation,149-152 step velocity profiles,103 the transition between growth mechanisms,80 and the effects of additives and impurities.135-137,145,146,153 AFM has also been used to distinguish between polymorphs for pharmaceutical compounds154 and to observe nucleation in proteins.11 Although AFM can be used in situ to obtain surface topography and step velocities, mass-transfer limitations at the crystal solution interface may persist, even at mass transfer Peclet numbers of O (106).8 These limitations result in the deviation of the solute concentration at the crystal surface from the inlet concentration and must be taken into consideration when determining kinetic coefficients from AFM observations. 6. Solvents, Impurities and Additives The importance of solvent selection in a crystallization process can be traced to the fact that the nature of the solvent-solute interactions at the crystal surface greatly influence the ultimate shape of the crystal. Wells drew attention to


these phenomena by citing solvent effects on the crystal shapes of iodoform and anthranilic acid155 and resorcinol155,156 (also see the work of Davey et al.157 for more on resorcinol). The crux of the matter is that a given crystallographic orientation on a crystal exposes a particular set of atoms of the solute molecule; therefore, different orientations have varying affinities for the solvent. These varying affinities impart different growth rates to the crystal faces and therefore influence the shape. Other examples of such behavior are abundant in the literature. For instance, Davey et al.158 demonstrated that succinic acid grown from water resembles a flat, diamond-like plate and exhibits large {001} basal planes bounded by {110} side faces, whereas succinic acid grown from isopropanol resembles an elongated hexagon with {001} basal planes bounded by the {110} and {010} side faces. Gordon and Amin159 disclosed that ibuprofen grown out of more polar solvents such as methanol or isopropanol yield crystals with a more equant aspect ratio, whereas ibuprofen crystals grown out of hexane yield much more-elongated crystals. Finally, Davey160 and Lahav and Leiserowitz161 have reviewed several other examples of solventinfluenced crystal shapes and discussed mechanisms related to the solvent effect. It is also well-known that foreign molecules (additives or impurities) influence the growth shape of crystals. These molecules can be divided into three categories: tailor-made additives, small-molecule additives, and macromolecules. Smallmolecule additives include surfactants and fatty acids. Assuming the crystallizing solute has a molecular weight of no more than 200-500 atomic mass units (amu), a macromolecular additive or impurity could be a protein, polymer, or peptide. An important type of molecular impurity or additive is one that is structurally related to the solute, sometimes referenced as being “tailor-made”. As the name suggests, a tailor-made additive is a molecule that is practically identical to the crystallizing solute molecule; these types of additives are relevant to crystallization, because they often occur as reaction byproduct formed during the synthesis of the solute. There are many reports in the literature that describe the effect of additives or impurities on theshapeorgrowthofcrystals,bothorganicandinorganic.137,162-165 Sections 6.1-6.3 describe some of these experiments, as well as past modeling efforts. 6.1. “Tailor-made” Additives. Weissbuch et al.166 demonstrated that chiral molecules can be directly assigned an absolute configuration (levorotary (L) or dextrorotary (D)), based on the changes that they caused to the morphology of a crystal composed of a prochiral molecule (i.e., an achiral molecule that can be converted to a chiral molecule in a single step) as the crystal is grown in the presence of either resolved or racemic molecular additives. In particular, the R polymorph of glycine (H3+NCH2COO-), when crystallized out of water, normally exhibits a coffin shape whose basal planes are the {010} family (i.e., the (010) and (01j0) planes). Figure 11a shows a schematic of the packing pattern of R-glycine, as viewed perpendicular to the b-direction; the molecules in “Position 1” are related to the molecules “Position 2” by a 180° rotation about the b-direction, together with a translation parallel to the b-direction, while the molecules in “Position 1” are related to the molecules in “Position 3” by an inversion operation. Replacing one of the two hydrogens on the central carbon atom of glycine with a methyl group yields alanine, but the prochiral nature of glycine dictates that the molecule obtained by such a replacement must yield either D-alanine or L-alanine, depending on which of the two hydrogens is replaced; either way, the resultant L- or D-alanine molecule is so similar to

Figure 11. (a) Packing arrangement of R-glycine, as viewed perpendicular to the b-direction. (b) Scheme illustrating the manner in which D-alanine is incorporated into the R-glycine lattice.

glycine that one might suppose, under the appropriate conditions, it could easily be incorporated into the R-glycine lattice. Figure 11b presents the R-glycine packing scheme into which a D-alanine molecule (circled) has been incorporated. Clearly, the amino and acid groups of D-alanine are in register with the rest of the crystal lattice, allowing D-alanine to be captured as if it were a glycine molecule, while its methyl group is oriented in an outward normal fashion to the (010) face. This methyl group can now interfere with the subsequent deposition of glycine molecules which, generally speaking, will decrease the growth rate of the (010) face. Moreover, this effect will not be observed on the (01j0) face, because of the chiral nature of D-alanine and the prochiral nature of glycine. If the molecules of “Position 1” or “Position 2” were to terminate at the (01j0) face, this would require that D-alanine be bound with its methyl group situated directly into the bulk of the crystalsthis is a very energetically unfavorable configuration and therefore unlikely to occur. If the molecules of “Position 3” or “Position 4” were to terminate at the (01j0) face, the methyl group of D-alanine would be required to be oriented perpendicular to the b-direction, again resulting in unfavorable interactions. Therefore, it is expected that D-alanine would interfere with growth on the (010) face and not the (01j0) face; the resulting asymmetry in growth rates should be manifested in a crystal with a much larger (010) face, relative to the (01j0) face. It is precisely these considerations that Weissbuch et al.166 used to anticipate and rationalize their results. Ultimately, these researchers observed that if D-alanine (CH3CH(NH2)COOH) was present in solution during crystallization, the R-glycine crystal


Figure 12. Growth shapes of crystals of (a-c) R-glycine166 and (d-f) adipic acid,167 as influenced by additives.

possessed a much larger (010) plane, as compared to the (01j0) face; in contrast, the crystal possessed a large (01j0) face if L-alanine were in solution. Their results are shown in Figure 12a-c. A U.S. patent by Klug and van Mil168 described the effect of properly chosen additives on the purity of adipic acid as it crystallized from water. They used both surfactants and tailormade additives and observed that certain additives caused a change in the normal shape of adipic acid such that crystals with the altered shape exhibited improved purity, with respect to the uptake of other structurally related molecules. In particular, the authors observed that when adipic acid, whose steady-state convex hull is bounded by the (dominant) {100}, {001}, and {011} forms, was crystallized from water in the presence of an amount on the order of 1 wt % (solvent-free basis) of any of glutaric acid (HOOC(CH2)3COOH), succinic acid (HOOC(CH2)2COOH), maleic acid (cis-HOOC(CH)2COOH) or fumaric acid (trans-HOOC(CH)2COOH). The resultant crystals were hardly distinguishable from the native crystal; that is, the {100} form still comprised most of the total area. However, when 3 wt % (solvent-free basis) of caproic acid (CH3(CH2)4COOH) was present, the adipic acid crystals exhibited a shape where the {001} form dominated. Klug and van Mil168 proposed that caproic acid, and no other additive, was incorporated into the {001} face, because of its similarity to adipic acid, and interfered with the subsequent addition of adipic acid molecules; moreover, they also observed a simultaneous reduction in the uptake of the other molecules. These experiments demonstrate that intelligently selected tailor-made additives can have a significant impact on the shape, which results from the specificity of interaction between the additive or impurity and both the solute molecule and its particular two-dimensional surface arrangement. For a more extensive and classical explanation of the effect of tailor-made additives and their effects on crystal growth, see the review by Weissbuch et al.169 As elegant as the tailormade additive approach is for manipulating crystal shape, it is worth mentioning that the tailor-made additive still must be incorporated into the lattice to exert its effect, thus short-circuiting the most basic purpose of crystallization, namely, separation and purification. In some instances, purity requirements for the solute in its crystalline form prohibit the use of tailor-made additives as an option to obtain a more-desirable shape. 6.2. Macromolecules. Michaels and Colville170 presented results from a systematic study on the morphology change

exerted by two surfactantssanionic sodium dodecylbenzene sulfonate (SDBS) and cationic trimethyldodecylammonium chloride (TMDAC)son adipic acid as it was crystallized from water. The authors observed that, when adipic acid was grown in the presence of 50 ppm SDBS, the {001} and {011} forms became increasingly dominant over the {100} form. At higher concentrations of SDBS (500 ppm), the {100} form had almost disappeared and the crystals resembled thin rods or needles with {001} and {011} side faces. On the other hand, when adipic acid was grown with 96 ppm TMDAC, the {100} form became larger, at the expense of the other forms. The authors determined that both SDBS and TMDAC reduced the growth rate of all forms, but the surfactants’ effects on the different forms were anisotropic. Indeed, the measured growth rates revealed that TMDAC was much more effective in reducing the growth rate of the {100} form, as compared to either the {011} or {001} form. Michaels and Tausch167 performed similar experiments, the results of which are reproduced here in Figure 12d-f. Another example of the effect of macromolecules on crystal growth has been observed in nature. Antarctic fish are able to survive in waters where temperatures fall as much as 1 °C below the equilibrium freezing point of their body’s fluids. Ice crystals are abundant in the waters at these temperatures. DeVries171 and Raymond and DeVries172 have demonstrated that these fish have glycoproteins and serum proteins, which prevent the growth of ice crystals in their blood. Unlike NaCl, these “antifreeze” proteins do not lower the freezing point through a colligative effect, as evidenced by the fact that the melting point of ice crystals in the fishes’ blood serum is virtually unchanged. Rather, the proteins preferentially adsorb onto nascent ice crystals and disrupt growth. 6.3. Modeling Impurity and Additive Effects. Cabrera and Vermilyea173 advanced the idea that flowing steps could be stopped by additives that adsorb onto a crystal surface and remain immobile (the term immobile is relative; more specifically, the additive must have an adsorption lifetime much longer than some characteristic time for growth). Their hypothesis was that flowing steps would be forced to percolate through a collection of additives; because the edge encountered the immobile additive, Cabrera and Vermilyea argued the edge would be “pinned” at the point of contact and a segment of edge between two additives would begin to curve as growth proceeded. If the spacing between the additives parallel to the edge direction was less than some critical length lc (see section 5.1.1), the steps would be partitioned into sections of less than critical length and growth would be arrested. If the additives were spaced a distance greater than lc, the step would pass between the additives and continue to grow; however, because of its curvature, Cabrera and Vermilyea argued the edge would experience a decrease in the driving force for growth, resulting in a diminished velocity. As shown in section 5.1.1, the critical radius of a two-dimensional nucleus is given as rc ≈ σ-1. Therefore, one consequence of Cabrera and Vermilyea’s model is that there will be some critical supersaturation σ* > 0 below which growth cannot occur (sometimes called a “dead zone”). Physically this means that, by decreasing the supersaturation to some nonzero value σ*, the critical length becomes larger than the spacing between the additives, and growth is arrested. Several accounts in the literature report the observation of this effect.146,174 Some additives may have adsorption lifetimes that are comparable to or shorter than the characteristic time for surface growth; in these cases, the aforementioned model does not hold. Because the velocity of a step is directly proportional to its kink


density (recall eq 28), a natural alternative is to consider the possibility that additives adsorb into kink sites and reduce their effective density. An equivalent statement is to consider that the additives increase the average distance between effective kink sites along an edge. Chernov175 and Cabrera and Coleman176 developed such an expression for a crystal growing from a vapor phase with additive partial pressure Pp by considering the relative equilibrium probabilities of finding each of the following along an edge: a straight segment, a kinked segment, a straight segment with an adsorbed additive, and an additive adsorbed at a kink site. If step velocity is proportional to kink density, (see section 5.1.3) as given by eq 28, then the (poisoned) kink density must relax to its equilibrium value on a time scale much shorter than some characteristic time scale for growth of the edge. Physically, this leads to the requirement that onedimensional particle exchange along the edge is much faster than two- or three-dimensional mass transport between the edge and the adlayer and/or the bulk phase. Clydesdale and co-workers177,178 extended the attachment energy model (see section 3) of crystal shape prediction71,179 by considering the effects of tailor-made additives on the attachment energies. Their approach was to “map” the atomic coordinates of a tailor-made additive molecule onto those of a host or solute molecule and calculate slice and attachment energies, both when the imposter was within and external to a particular layer. Modified attachment energies were implemented in a shape model to yield shapes different from those predicted by the native model. The researchers used a quantity that was developed by Berkovitch-Yellin180 called the “binding energy” (∆b), which allowed them to determine which faces would favorably receive imposter molecules. The binding energy compares the energies of a host molecule after being bound up in a layer, Eslice, and subsequent attachment of the layer to a substrate, Eatt(-), to those of an imposter molecule bound up in ′ the same layer, Eslice , and subsequent attachment of the layer to ′ ′ ′ a substrate, Eatt(-) . Defining ∆b ≡ Eslice + Eatt(-) - (Eslice + Eatt(-)), values of ∆b > 0 imply unfavorable imposter binding, values of ∆b < 0 imply favorable imposter binding, and values of ∆b ) 0 imply indifferent imposter binding. The step-pinning model of Cabrera and Vermilyea173 and the kink density reduction model of Chernov175 and Cabrera and Coleman176 address the problem of predicting the effect of additives or impurities on crystal growth by considering which mechanisms must be at work. In the case of the step-pinning model, one must know the extent to which additives or impurities have adsorbed on a crystal face. The kink density reduction model requires estimating the chemical potential of the additive or impurity in a kink site. The modified attachment energy model of Clydesdale et al.,177,178 although helpful for characterizing crystal faces with respect to their ability to receive tailor-made additives or impurities in a heuristic fashion, does little to elucidate exactly how the growth rate of a face is reduced. A common approach to modeling the effects of additives or impurities is to (1) assume a mechanism (such as those described previously), (2) assume the form of an adsorption isotherm, (3) conduct experiments, and (4) regress the data to estimate all the unknown parameters in the model equations. While this approach may correlate the measured data in a satisfactory manner, it is unlikely to result in a predictive methodology. A first-principles model is more desirable, which would predict everything, from the level of surface adsorption of the additives or impurities to the reduced growth rate and also account for any kind of growth-reduction mechanism. Such a model can only spring from an understanding of the processes

that govern the movements of molecules in and out of the surface during crystallization. 7. Crystal Shape Evolution Model Whether growing from a pure solution or in the presence of additives or impurities, crystals can eventually reach a steadystate shape; that is, the shape of the crystal has the property of being self-similar, even as it continues to grow. However, in a crystallizer, crystals do not all necessarily attain their steadystate shape, nor does the shape of a crystal remain constant during growth. Consequently, it is important to have a method that predicts the evolving shape of a crystal grown from an arbitrary starting shape (e.g., seed) at any subsequent moment under realistic operating conditions. This method, under the appropriate limiting conditions, should also show that a crystal possesses a steady-state shape. 7.1. Geometric View of Crystal Growth. A geometric view of surface evolution means that the outward normal velocity of a surface is dependent only on its local orientation. The physical implication is that each point on the surface evolves in only one direction (forward), unlike, for instance, diffusive phenomena. The governing equation for geometric surface evolution is the Hamilton-Jacobi equation; the solution(s) to this equation describe(s) many types of phenomena,181-183 including the growth of crystals.64 Solutions to the Hamilton-Jacobi equation exhibit so-called “shock transitions”, which, for crystal growth, are interpreted as transitions from smooth growth to faceted growth. However, by assuming that crystals are faceted for all times, Zhang et al.66 developed a more straightforward and readily applicable evolution model as follows. Let Ghkl be the perpendicular growth rate of the hkl face; it follows that the rate of change of the perpendicular distance (Hhkl) from an origin inside the crystal to the face is given by dHhkl(t) ) Ghkl (33) dt where this equation applies to each of N faces. Zhang et al.66 nondimensionalized eq 33 by selecting a reference face and using its corresponding values of Href and Gref to define the relative perpendicular distance (xhkl ≡ Hhkl/Href) and the relative growth rate (Rhkl ≡ Ghkl/Gref). Coupled with a dimensionless time dξ ) (Gref/Href) dt, they derived the following dynamical model for the remaining N - 1 faces, assuming that the crystal is faceted for all times: dxhkl(ξ) ) Rhkl - xhkl dξ

(34)

where

( )

ξ ) ln

Href(t) Href(0)

(35)

When all of the Rhkl values are constant, eq 34 clearly displays a unique, stable steady-state value, xˆhkl, which is given by xˆhkl ) Rhkl

(36)

Equation 36 is equivalent to the Frank-Chernov condition64,65 (see section 3). A consequence of this steady-state condition is that the steady-state growth shape of a crystal can be predicted by knowing (either by measurement or prediction) the relative growth rates of each face. Conversely, the relative growth rates of the steady-state faces can be obtained from the steady-state growth shape. The growth models in section 5 can be used to develop expressions for each Rhkl and eq 34 can be integrated


Figure 13. (a-d) Predicted evolution of ibuprofen (∆ξ ) 0.025) and (e) experimental steady-state shape (not drawn to scale). Reprinted (e) with permission from the work of cano et al.185 2001 Elsevier.

to determine how the shape of a seed crystal (or a nucleus) evolves with time. 7.2. Evolution Case Study: Ibuprofen. Ibuprofen is a commonly used anti-inflammatory drug. Its crystal structure (as reported by Shankland et al.184) belongs to the monoclinic P21/c space group, with the following lattice parameters: a ) 14.397 Å, b ) 7.818 Å, c ) 10.506 Å, and β ) 99.70°. Its unit cell is comprised of four molecules. Following the method of Winn and Doherty,89 the {100}, {011}, and {002} forms were determined to grow from an ethanol solution via the screw dislocation mechanism, with relative perpendicular growth rates of Rhkl ) 1.00, 2.67, and 1.45, respectively. Other low-index forms (e.g., {111j} or {102j}) and higher-index forms were assigned relative growth rates of 10, because they contained no stable edges and therefore are expected to grow via the rough growth mechanism, which, as noted in section 4.1 is expected to occur at a rate at least 10 times faster than growth via a layered mechanism. The aforementioned relative growth rates account for the effect of the solvent. To show the evolution of ibuprofen from an arbitrary state, a “spherical” seed was used as the initial condition for eq 34. This seed was composed of 102 faces situated equidistant from the origin (see Figure 13a). The crystal was then evolved from this seed, assuming constant relative growth rates for all times ξ and discretizing eq 34 with a time step of ∆ξ ) 0.025. The results of this evolution are shown in Figures 13a-d. As can be seen, the faster-growing forms disappear, leaving only the {100}, {011}, and {002} forms on the steady-state growth shape. The predicted steady-state shape as ξ f ∞ is shown in Figure 13d. It has a large hexagonal {100} form that is slightly elongated in the b-direction, because R011 > R002. An experimental shape of ibuprofen grown from ethanol obtained by Cano et al.185 is shown in Figure 13e, for comparison with the prediction. It is important to recognize that the shape in Figure 13d is a pure prediction and nothing about the measured shape was used to obtain it; the only experimental data used in this prediction were the single-crystal X-ray data which furnished the threedimensional atomic coordinates, the crystal space group, the unit-cell dimensions, and the solvent solubility parameter data. This example also demonstrates an important rule of thumb for crystal shape: faces with large perpendicular growth rates will either possess a small area on the steady-state shape or disappear altogether. The ultimate area of each face on the steady-state shape is dependent not only on the relative growth rates of each face but also on the angles between each of the neighboring faces. 8. Coupling Size and Shape Predictions So far, we have described experiments, simulations, and models for the shape prediction and evolution of a single crystal growing from solution. However, when considering the design

and control of industrial-scale crystallizers, the shape of the entire population of crystals becomes more relevant. There is often a particular interest in achieving a target particle size and shape distribution (PSD) of the crystal population, which is modeled using population balance equations (PBEs).186,187 8.1. Multi-Dimensional Population Balances. To track the distribution of crystal morphologies inside a crystallizer, shape effects must be incorporated into PSD modeling. Iveson188 suggested that multidimensional population balances must be used, because of the limitations of conventional singledimensional PBEs that consider particle size in only one internal direction. For multidimensional population balances, a growth rate is needed for every independent direction that is used in the model; techniques for measuring growth rates in multiple directions are discussed in the following sections. Several studies have developed 2D PBEs to track two characteristic crystal dimensions (usually length and width). Ma et al.189,190 used a compartmental model of a batch crystallizer to account for imperfect mixing, coupled with a 2D PBE, to model the growth of needlelike potassium dihydrogen phosphate (KDP) crystals. For continuous crystallization, Puel and co-workers191-193 incorporated a detailed kinetic study of growth rates, primary and secondary nucleation, and diffusion effects into a 2D PBE model, and they were able to predict the shape evolution of the rodlike hydroquinone crystals accurately. Oullion et al.194,195 studied the batch cooling crystallization of plateletlike organic crystals and included growth and secondary nucleation mechanisms into their 2D PBE to determine the effects of seeding and cooling profiles on the final PSD. Other researchers have demonstrated the use of particle properties other than characteristic length in their multidimensional population balances. For instance, Gerstlauer et al.196 considered the internal lattice strain of individual crystals as a particle coordinate, and Briesen197 used a coordinate transformation on the multidimensional PBE to model the system by the crystal volume and a shape factor. Briesen demonstrated his particle coordinate transformation technique by simulating a batch cooling crystallizer and inverting the transformation to obtain the original 2D number density function at the initial and final times, as shown in Figure 14a. A different approach that was developed by Zhang and Doherty198 coupled variations in the shape evolution model described in section 7 with a one-dimensional (1D) population balance to predict crystal shape evolution and size distribution simultaneously. This method is applicable to crystals that can be treated as being essentially two-dimensional (e.g., plates) and assumes that the relative growth rates of crystal faces are constant. This model requires only one absolute growth rate (for a reference face); all others are found knowing the relative growth rates of all other faces, relative to the reference face. This approach gives the size and shape evolution of every


Figure 14. Examples of multidimensional population balance equations (PBEs): (a) Two-dimensional representation, with respect to the characteristic lengths of the particle number density function at the initial and final times. Reprinted with permission from the work of Briesen.197 2006 Elsevier. (b) Steady-state crystal size and shape distribution for triangular seeds of succinic acid grown from water. Reprinted with permission from the work of Zhang and Doherty.198 2004 American Institute of Chemical Engineers.

independent face. The method was applied to succinic acid grown from water, as shown in Figure 14b. Ma et al.199 have recently used polyhedral PBE modeling to predict the PSD for every independent crystal face. This model uses knowledge of the crystal shape of a single crystal, as well as the measured growth rates for each independent family of planes; there are three sets of these planes in their study of potash alum. This prediction of individual facet growth at different times leads to the simultaneous prediction of crystal shape and size evolution. One difficulty in implementing multidimensional PBEs is the increased computational efforts required to solve the models. The studies discussed above used a variety of techniques to overcome these issues, including high-resolution simulation algorithms that coupled upwind discretization and the LaxWendroff method,189,190,200,201 the method of classes,191-193,195,199 model reduction by means of a coordinate transformation method,197 and other discretization methods (such as the method of lines196 and the finite difference method198). Multidimensional population balances also appear in the modeling of many particulate processes other than crystallization, such as granulation, polymerization, and cell dynamics, which have led to new solution techniques for these numerically intensive models. Population balance discretization, using a hierarchial two-tier solution strategy,202 adaptive high-resolution finite volume schemes203 and a discretization method based on a generalized moving pivot technique204 are a few of the recent methods developed for solving multidimensional PBEs. 8.2. Particle Size and Shape Characterization. The measurement and characterization of particle size and shape distributions are important for implementing and validating the methods that have been described previously for determining PSDs. Several techniques and sensors are available that can measure particle size and/or PSDs (such as infrared spectroscopy and laser diffraction) but they generally cannot be used to measure particle shape directly. The most widely used technique today for simultaneous particle shape and size characterization is in situ monitoring via video microscopy coupled with image analysis. Various in situ monitoring tools currently in use include Lasentec’sFocusedBeamReflectanceMeasurement(FBRM),205-208 which provides the chord length distribution (CLD) of the crystal population (many algorithms are currently available for transforming a CLD into a PSD), Lasentec’s Process Vision and

Measurement (PVM)205,207-209 that can take two-dimensional images of specific portions of the crystal population and the online imaging system by GlaxoSmithKline209-217 among other online and off-line imaging devices. These tools can be used in conjunction with conventional in situ techniques, such as ATR-FTIR spectroscopy205,207,218 which is used for solution concentration measurements to measure supersaturation, and Raman spectroscopy,207 which is used for polymorph monitoring. Commercial image analysis software is available but generally fails to segment crystals from the image successfully, because of the crystallizer hydrodynamics and continuous particle movement,219 thus motivating the need for improved online image analysis techniques. One method to evaluate crystal size and shape information from in situ crystallization images is via image segmentation, which separates crystals from the image background. Puel and coworkers have developed off-line imaging techniques to characterize crystal shape and size in two dimensions;191,193,194 however, these methods cannot be easily implemented in situ. Calderon de Anda et al.209 presented a multiscale segmentation method that takes into account the varied pixel intensity of online images (this article also provides a comprehensive review of other image segmentation techniques that were available at that time). Larsen et al. developed a commonly used image analysis algorithm, using a model-based object recognition for needlelike particles,216 and, more recently, extended the method to crystals of any shape that can be represented by a wire-frame model;217 the authors were able to successfully implement these methods on solutions of moderate solids concentration. Figure 15 shows examples of the model-based shape recognition for crystals (M-SHARC) algorithm that was developed by Larsen et al.217 for solutions with low solids concentration (Figures 14a and c) and high solids concentration (Figures 14b and d). Solutions of high solids density generally present a challenge to image segmentation methods220 because of the overlap of crystals and other phenomena, such as agglomeration, breakage, and attrition. After crystal images are segmented from the image background, size and shape descriptors are used to characterize possible crystal shapes. Recent work on shape descriptors include contour-based and region-based methods to determine aspect ratios, convexity, roundness, as well as other shape and


Figure 15. Image segmentation results for solutions of (a and c) low solids concentration and (b and d) high solids concentration. (Reproduced with permission from Larsen et al.217 Copyright Elsevier, 2007).

size properties. A single shape descriptor is not sufficient to distinguish crystals of different shapes prompting the use of alternative shape descriptors.219,221-223 A recent review by Zhang and Lu224 discussed several shape descriptor techniques. Another commonly used technique involves the application of Fourier descriptors to transform contours and regions of particle images into shape characteristics.209,213,225 Most image segmentation and shape descriptor techniques provide 2D shape information, whereas a recently developed camera model by Li et al.213 constructs 3D shapes from 2D images. The online imaging and image analysis techniques previously discussed can be used to estimate crystal face growth rates and kinetics. One approach is to model crystal growth and shape using multidimensional population balances (as discussed previously) and estimate kinetic parameters from off-line images;205 however, to implement these measurements into a control scheme, real-time growth kinetics are required. Puel and coworkers measured the in situ growth kinetics of rodlike hydroquinone crystals193 and later developed a two-dimensional model to simulate the growth of platelet-shaped crystals in batchseeded cooling crystallization, to determine the kinetics of nucleation and attrition mechanisms.195 L-Glutamic acid has multiple polymorphs and the needlelike β form has been the subject of several in situ imaging studies modeled by multidimensional population balances to estimate growth kinetics.207,214,215 Kempkes et al.208 presented a procedure to interpret in situ images to estimate growth rates of a population of cuboids. The experimental studies presented here are generally limited to measuring growth rates in two directions, because of the 2D imaging techniques that are currently available; however, in the future, with proper 3D imaging, the growth rates of individual facets could be measured.226 8.3. Shape Control. Population balance models, coupled with online measurements, can be used to develop modelpredictive algorithms. Such algorithms can provide insight into optimal control policies to produce crystals of a certain size and shape. Patience and Rawlings219 used in situ images from a cooling crystallizer to develop a feedback control algorithm, using additive concentration as the manipulated variable to control crystal shape. They effectively added sodium dithionate

to a solution of growing sodium chlorate crystals, to change their shapes from cubic to tetrahedral, while using the percentage of cubic crystals as the input to the feedback controller. In another study by Yang et al.,227 the effects of different cooling modes were investigated for a batch cooling crystallization, using the supersaturation as the manipulated variable to control the shape of KDP crystals; this can be beneficial in selecting an optimal cooling profile for a desired product shape. Ultimately, if shape and size can be effectively controlled, then there is also the possibility of controlling polymorphic transformations, as demonstrated by the C-control strategies implemented by Hermanto et al.228 for the transformation of L-glutamic acid from the metastable R form to the stable β form. Braatz and co-workers229 recently reviewed advances in the direct design and control of pharmaceutical crystallization processes and discussed the possible applications to more-complicated tasks, such as polymorph control and protein crystallization. Through the use of multidimensional population balance models, in situ imaging techniques, and real-time image analysis, model-predictive control schemes could potentially automatically control the size and shape of an entire crystal population to meet product specifications. For further insight into the techniques presented in this section, the reader is referred to recent reviews in refs 226, 230, and 231. 9. Conclusions and Future Directions Significant progress has been made over the past several years, with regard to developing engineering tools that allow for improved crystallization processing design. In particular, methods have been developed to predict steady-state crystal shapes through both molecular simulation and mechanistic modeling. Some of these methods provide the ability to account for the effects of single solvents as well as tailor-made additives on crystal shapes. The mechanistic modeling techniques can be performed fast enough (on the order of minutes of calculation time for current personal computers) that they have the potential to be implemented as part of product and process development. As we have shown through scaling arguments, the shapes of most molecular organic crystals growing by spirals are determined by the rates of surface integration on each face and not by bulk transport limitations. Thus, bulk transport effects in solution growth of molecular organic crystals are important only insofar as the local supersaturation is a function of position within the crystallizer. Further development continues for multidimensional population balances, including their solution techniques and implementation in real-time feedback control. The combination of these modeling efforts with online shape measurement and image analysis techniques provides a method to measure real-time growth kinetics, indicating a promising future for model-predictive crystallization control. In the future, continued research in these areas should allow for the prediction of (i) crystal shapes, as a function of supersaturation; (ii) the appropriate dominant growth mechanism for a given set of processing conditions; and (iii) the effect of multiple solvents and various types of additives on crystal shape. These predictive capabilities will enable the process design stage of crystalline product development to include, for instance, both solubility and shape considerations in solvent selection. Furthermore, a priori shape predictions based on mechanistic growth models (described in section 5.1) will be developed for inorganic material as new techniques for modeling long-range interactions and simultaneous reaction and crystallization are developed. In addition, these shape predictions can be coupled with devices that track the growth rates of individual crystal faces, improving


the ability to modify crystal shapes and particle size and shape distributions (PSDs) in real time during crystallizer operation. Several aspects of crystal shape engineering will require the development of breakthrough technologies for the field to continue to advance. To couple polymorph and solvent selection, new (more reliable) methods are required for polymorph prediction. Currently, a gap exists in the knowledge of transitions between the growth mechanisms of nuclei (where the polymorphic structure is established) to those of faceted crystals. Filling this knowledge gap will require significant improvements in simulation and experimental techniques to handle the range of length and time scales involved. While the relative growth rates of crystal faces can be predicted for several systems, a major advance is necessary to predict absolute growth rates from first principles. Finally, the broader implementation of crystal shape engineering as a practice in product and process design will require the development of computer-aided design tools for polymorph selection and shape prediction that are integrated with thermodynamic calculations (e.g., solubilities and surface energies) and databases (e.g., crystallography, solvent properties, etc.). Acknowledgment The authors are grateful for financial support provided by the National Science Foundation (through Grant Nos. CTS0554718 and CBET-0651711). Acknowledgment is also made to Merck and Co., the Donors of the American Chemical Society Petroleum Research Fund (under Grant No. 44172-AC10), and the International Fine Particle Research Institute (IFPRI) for partial support of this research. The work of A.R.B. was supported by graduate fellowships from the National Science Foundation and the UC Regents. The authors also thank Prof. Todd Squires for helpful discussions of transport phenomena. Supporting Information Available: Enclosed in the Supporting Information is a derivation of the bulk-transport-limited model and a description of the method used to obtain the scaling of the depletion boundary layer thickness. (PDF) This material is available free of charge via the Internet at http://pubs.acs.org. Literature Cited (1) Yin, J.-C.; Zhou, J.-S.; Sun, J.; Qiu, Y.; Wei, D.-Z.; Shen, Y.-L. Study of the crystal shape and its influence on the anti-tumor activity of tumor necrosis factor-related apoptosis-inducing ligand (Apo2L/TRAIL). Cryst. Res. Technol. 2008, 43, 888. (2) Yang, H. G.; Sun, C. H.; Qiao, S. Z.; Zou, J.; Liu, G.; Smith, S. C.; Cheng, H. M.; Lu, G. Q. Anatase TiO2 single crystals with a large percentage of reactive facets. Nature 2008, 453, 638. (3) Prestidge, C. A.; Tsatouhas, G. Wettability studies of morphine sulfate powders. Int. J. Pharm. 2000, 198, 201. (4) Heng, J.; Williams, D. Wettability of Paracetamol Polymorphic Forms I and II. Langmuir 2006, 22, 6905. (5) Heng, J.; Bismarck, A.; Lee, A.; Wilson, K.; Williams, D. Anisotropic Surface Energetics and Wettability of Macroscopic Form I Paracetamol Crystals. Langmuir 2006, 22, 2760. (6) Riebel, U.; Kofler, V.; Lffler, F. Shape Characterization of Crystals and Crystal Agglomerates. Part. Part. Syst. Charact. 1991, 8, 48. (7) Variankaval, N.; Cote, A. S.; Doherty, M. F. From Form to Function: Crystallization of Active Pharmaceutical Ingredients. AIChE J. 2008, 54, 1682. (8) Law, M.; Greene, L. E.; Johnson, J. C.; Saykally, R.; Yang, P. Nanowire dye-sensitized solar cells. Nat. Mater. 2005, 4, 455. (9) Bauer, J.; Spanton, S.; Henry, R.; Quick, J.; Dziki, W.; Porter, W.; Morris, J. Ritonavir: An Extraordinary Example of Conformational Polymorphism. Pharm. Res. 2001, 18, 859. (10) Bernstein, J. Polymorphism in Molecular Crystals; Clarendon Press: Oxford, U.K., 2002.

(11) Yau, S.; Vekilov, P. Direct Observation of Nucleus Structure and Nucleation Pathways in Apoferritin Crystallization. J. Am. Chem. Soc. 2001, 123, 1080. (12) Yu, L. Nucleation of One Polymorph by Another. J. Am. Chem. Soc. 2003, 125, 6380. (13) Gibbs, J. W. The Scientific Papers of J. Willard Gibbs, Vol. 1, Thermodynamics; Dover: New York, 1961. (14) Volmer, M.; Weber, A. Z. Phys. Chem. 1926, 119, 227. (15) Farkas, Z. Z. Phys. Chem. (Leipzig) 1927, 125, 236. (16) Becker, R.; Do¨ring, W. Ann. Phys. 1935, 24, 719. (17) Kelton, K. F. Crystal Nucleation in Liquids and Glasses; Solid State Physics, Vol. 45; Academic Press: New York, 1991. (18) Zettlemoyer, A., Ed. Nucleation; Marcel Dekker: New York, 1969. (19) Kashchiev, D. Nucleation: Basic Theory with Applications; Butterworth-Heinemann: Boston, MA, 2000. (20) Wu, D.; Granasy, L.; Spaepen, F. Nucleation and the Solid-Liquid Interfacial Free Energy. MRS Bull. 2004, (Dec.), 945. (21) Zykova-Timan, T.; Valeriani, C.; Sanz, E.; Frenkel, D.; Tosatti, E. Irreducible Finite-Size Effects in the Surface Free Energy of NaCl Crystals from Crystal-Nucleation Data. Phys. ReV. Lett. 2008, 100, 036103. (22) Leyssale, J.-M.; Delhommelle, J.; Millot, C. Hit and miss of classical nucleation theory as revealed by a molecular simulation study of crystal nucleation in supercooled sulfur hexafluoride. J. Chem. Phys. 2007, 127, 044504. (23) Cacciuto, A.; Frenkel, D. Stresses Inside Critical Nuclei. J. Phys. Chem. B 2005, 109, 6587. (24) Kashchiev, D.; Vekilov, P. G.; Kolomeisky, A. B. Kinetics of twostep nucleation of crystals. J. Chem. Phys. 2005, 122, 244706. (25) ten Wolde, P. R.; Frenkel, D. Enhancement of Protein Crystal Nucleation by Critical Density Fluctuations. Science 1997, 277, 1975. (26) Lomakin, A.; Asherie, N.; Benedek, G. Liquid-solid Transition in Nuclei of Protein Crystals. Proc. Natl. Acad. Sci., U.S.A. 2003, 100, 10254. (27) Soga, K. G.; Melrose, J. R.; Ball, R. C. Metastable states and the kinetics of colloid phase separation. J. Chem. Phys. 1999, 110, 2280. (28) Sayle, T.; Catlow, C.; Maphanga, R.; Ngoepe, P.; Sayle, D. Generating MnO2 Nanoparticles Using Simulated Amorphization and Recrystallization. J. Am. Chem. Soc. 2005, 127, 12828. (29) Vekilov, P. Dense Liquid Precursor for the Nucleation of Ordered Solid Phases from Solution. Cryst. Growth Des. 2004, 4, 671. (30) Pan, W.; Kolomeisky, A. B.; Vekilov, P. G. Nucleation of ordered solid phases of proteins via a disordered high-density state: Phenomenological approach. J. Chem. Phys. 2005, 122, 174905. (31) Zhang, T.; Liu, X. How Does a Transient Amorphous Precursor Template Crystallization. J. Am. Chem. Soc. 2007, 129, 13520. (32) van Duijneveldt, J. S.; Frenkel, D. Computer simulation study of free energy barriers in crystal nucleation. J. Chem. Phys. 1992, 96, 4655. (33) ten Wolde, P. R.; Ruiz-Montero, M. J.; Frenkel, D. Numerical Evidence for bcc Ordering at the Surface of a Critical fcc Nucleus. Phys. ReV. Lett. 1995, 75, 2714. (34) Valeriani, C.; Sanz, E.; Frenkel, D. Rate of homogeneous crystal nucleation in molten NaCl. J. Chem. Phys. 2005, 122, 194501. (35) Gasser, U.; Weeks, E.; Schofield, A.; Pusey, P.; Weitz, D. RealSpace Imaging of Nucleation and Growth in Colloidal Crystallization. Science 2001, 292, 258. (36) Balsara, N. P.; Rappl, T. J.; Lefebvre, A. A. Does conventional nucleation occur during phase separation in polymer blends. J. Polym. Sci., Part B: Polym. Phys. 2004, 42, 1793. (37) Pino-Garcia, O.; Rasmuson, A. Influence of Additives on Nucleation of Vanillin: Experiments and Introductory Molecular Simulations. Cryst. Growth Des. 2004, 4, 1025. (38) Roelands, C.; Roestenberg, R.; ter Horst, J.; Kramer, H.; Jansens, P. Development of an Experimental Method to Measure Nucleation Rates in Reactive Precipitation. Cryst. Growth Des. 2004, 4, 921. (39) Towler, C.; Davey, R.; Lancaster, R.; Price, C. Impact of Molecular Speciation on Crystal Nucleation in Polymorphic Systems. J. Am. Chem. Soc. 2004, 126, 13347. (40) Weissbuch, I.; Lahav, M.; Leiserowitz, L. Toward Stereochemical Control, Monitoring, and Understanding of Crystal Nucleation. Cryst. Growth Des. 2003, 3, 125. (41) Meijer, E.; Frenkel, D. Melting Line of Yukawa System by Computer Simulation. J. Chem. Phys. 1991, 94, 2269. (42) O’Malley, B.; Snook, I. Crystal Nucleation in the Hard Sphere System. Phys. ReV. Lett. 2003, 90, 085702. (43) Anwar, J.; Boateng, P. Computer Simulation of Crystallization from Solution. J. Am. Chem. Soc. 1998, 120, 9600. (44) Auer, S.; Frenkel, D. Numerical prediction of absolute crystallization rates in hard-sphere colloids. J. Chem. Phys. 2004, 120, 3015.

9830 Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 (45) Trudu, F.; Donadio, D.; Parrinello, M. Freezing of a Lennard-Jones Fluid: From Nucleation to Spinodal Regime. Phys. ReV. Lett. 2006, 97, 105701. (46) Desgranges, C.; Delhommelle, J. Polymorph selection during the crystallization of Yukawa systems. J. Chem. Phys. 2007, 126, 054501. (47) Browning, A. R.; Doherty, M. F.; Fredrickson, G. H. Nucleation and polymorph selection in a model colloidal fluid. Phys. ReV. E: Stat., Nonlinear, Soft Matter Phys. 2008, 77, 041604. (48) Beaucage, P.; Mousseau, N. Nucleation and crystallization process of silicon using the Stillinger-Weber potential. Phys. ReV. B: Condens. Matter Mater. Phys. 2005, 71, 094102. (49) Desgranges, C.; Delhommelle, J. Molecular Insight into the Pathway to Crystallization of Aluminum. J. Am. Chem. Soc. 2007, 129, 7012. (50) Ghiringhelli, L. M.; Valeriani, C.; Meijer, E. J.; Frenkel, D. Local Structure of Liquid Carbon Controls Diamond Nucleation. Phys. ReV. Lett. 2007, 99, 055702. (51) Desgranges, C.; Delhommelle, J. Insights into the Molecular Mechanism Underlying Polymorph Selection. J. Am. Chem. Soc. 2006, 128, 15104. (52) Desgranges, C.; Delhommelle, J. Controlling Polymorphism during the Crystallization of an Atomic Fluid. Phys. ReV. Lett. 2007, 98, 235502. (53) Radhakrishnan, R.; Trout, B. Nucleation of Hexagonal Ice (Ih) in Liquid Water. J. Am. Chem. Soc. 2003, 125, 7743. (54) Izumi, S.; Hara, S.; Kumagai, T.; Sakai, S. Molecular dynamics study of homogeneous crystal nucleation in amorphous silicon. J. Cryst. Growth 2005, 274, 47. (55) Leyssale, J. M.; Delhommelle, J.; Millot, C. A molecular dynamics study of homogeneous crystal nucleation in liquid nitrogen. Chem. Phys. Lett. 2003, 375, 612. (56) Zahn, D. Atomistic Mechanism of NaCl Nucleation from an Aqueous Solution. Phys. ReV. Lett. 2004, 92, 040801. (57) Gavezzotti, A. Molecular Aggregation of Acetic Acid in a Carbon Tetrachloride Solution: A Molecular Dynamics Study with a View to Crystal Nucleation. Chem. Eur. J. 1999, 5, 567. (58) Beckham, G.; Peters, B.; Starbuck, C.; Variankaval, N.; Trout, B. Surface-Mediated Nucleation in the Solid-State Polymorph Transformation of Terephthalic Acid. J. Am. Chem. Soc. 2007, 129, 4714. (59) Peters, B.; Trout, B. L. Obtaining reaction coordinates by likelihood maximization. J. Chem. Phys. 2006, 125, 054108. (60) Wulff, G. Z. Kryst. 1901, 34, 449. (61) Herring, C. Some Theorems on the Free Energies of Crystal Surfaces. Phys. ReV. 1951, 82, 87. (62) Mutaftschiev, B. Nucleation Theory. In Handbook of Crystal Growth, Vol. 1a, FundamentalssThermodynamics and Kinetics; Hurle, D. T. J., Ed.; North-Holland: Amsterdam, 1993; Chapter 4, pp 187-247. (63) Kern, R. The Equilibrium Form of a Crystal. In Morphology of Crystals: Part A; Sunagawa, I., Ed.; Terra Scientific Publishing Company: Tokyo, 1987; Chapter 2, pp 77-206. (64) Frank, F. C. On the Kinematic Theory of Crystal Growth and Dissolution Processes. In Growth and Perfection of Crystals; Doremus, R. H., Roberts, B. W., Turnbull, D., Eds.; Wiley: New York, 1958. (65) Chernov, A. A. The Kinetics of the Growth Forms of Crystals. SoV. Phys. Cryst. 1963, 7, 728. (66) Zhang, Y.; Sizemore, J. P.; Doherty, M. F. Shape evolution of 3-dimensional faceted crystals. AIChE J. 2006, 52, 1906. (67) Snyder, R. C.; Doherty, M. F. Faceted crystal shape evolution during dissolution or growth. AIChE J. 2007, 53, 1337. (68) Bravais, A. E´tudes Crystallographiques; Gauthier-Villars: Paris, 1866. (69) Friedel, M. G. E´tudes Sur la loi de Bravais. Bull. Soc. Fr. Miner. 1907, 9, 326. (70) Donnay, J. D. H.; Harker, D. A New Law of Crystal Morphology Extending the Law of Bravais. Am. Mineral. 1937, 22, 446. (71) Hartman, P.; Perdok, W. G. On the relations between structure and morphology of crystals. I. Acta Crystallogr. 1955, 8, 49. (72) Hartman, P.; Perdok, W. G. On the relations between structure and morphology of crystals. II. Acta Crystallogr. 1955, 8, 521. (73) Clydesdale, G.; Docherty, R.; Roberts, K. J. HABITsA program for predicting the morphology of molecular crystals. Comput. Phys. Commun. 1991, 64, 311. (74) Grimbergen, R. F. P.; Meekes, H.; Bennema, P.; Strom, C. S.; Vogels, L. J. P. On the Prediction of Crystal Morphology. I. The HartmanPerdok Theory Revisited. Acta Crystallogr., Sect. A: Found. Crystallogr. 1998, 54, 491. (75) Hartman, P.; Bennema, P. The attachment energy as a habit controlling factor: I. Theoretical considerations. J. Cryst. Growth 1980, 49, 145. (76) Kossel, W. Nachur. Ges. Gottingen 1927, 206, 135. (77) Stranski, I. N. Z. Phys. Chem. 1928, 136, 259.

(78) Kuroda, T. Growth of a crystal surface with non-uniformity in supersaturation due to laminar flow of solution along the surface. J. Cryst. Growth 1985, 71, 84. (79) Wilcox, W. R. Transport phenomena in crystal growth from solution. Prog. Cryst. Growth Charact. Mater. 1993, 26, 153. (80) Land, T. A.; De Yoreo, J. J. The evolution of growth modes and activity of growth sources on canavalin investigated by in situ atomic force microscopy. J. Cryst. Growth 2000, 208, 623. (81) Chernov, A. A. Modern Crystallography III. Crystal Growth; Springer-Verlag: Berlin, 1984. (82) Frenkel, J. On the surface motion of particles in crystals and the natural roughness of crystalline faces. J. Phys. U.S.S.R. 1945, 9, 392. (83) Burton, W. K.; Cabrera, N.; Frank, F. C. The Growth of Crystals and the Equilibrium Structure of Their Surfaces. Philos. Trans. R. Soc. A 1951, 243, 299. (84) Zhang, J.; Nancollas, G. H. Kink densities along a crystal surface step at low temperatures and under nonequilibrium conditions. J. Cryst. Growth 1990, 106, 181. (85) Cuppen, H. M.; Meekes, H.; van Veenendaal, E.; van Enckevort, W. J. P.; Bennema, P.; Reedijk, M. F.; Arsic, J.; Vlieg, E. Kink density and propagation velocity of the [010] step on the Kossel (100) surface. Surf. Sci. 2002, 506, 183. (86) Cuppen, H. M.; Meekes, H.; van Enckevort, W. J. P.; Vlieg, E.; Knops, H. J. F. Nonequilibrium free energy and kinetic roughening of steps on the Kossel (001) surface. Phys. ReV. B: Condens. Matter Mater. Phys. 2004, 69, 245404. (87) Volmer, M. Z. Phys. 1922, 9, 193. (88) Ohara, M.; Reid, R. C. Modeling Crystal Growth Rates from Solution; Prentice-Hall: Englewood Cliffs, NJ, 1973. (89) Winn, D.; Doherty, M. F. A new technique for predicting the shape of solution-grown organic crystals. AIChE J. 1998, 44, 2501. (90) Cuppen, H.; Meekes, H.; van Enckevort, W.; Vlieg, E. Birth-andspread growth on the Kossel and a non-Kossel surface. J. Cryst. Growth 2006, 286, 188. (91) De Yoreo, J. J.; Vekilov, P. G. Principles of Crystal Nucleation and Growth. In Biomineralization; Dove, P. M., De Yoreo, J. J., Weiner, S., Eds.; Reviews in Mineralogy and Geochemistry, Vol. 54; Mineralogical Society of America: Washington, DC, 2003. (92) Volmer, M.; Schultze, W. Z. Phys. Chem. A 1931, 156, 1. (93) Frank, F. C. The influence of dislocations on crystal growth. Disc. Faraday Soc. 1949, 5, 48. (94) Snyder, R. C.; Doherty, M. F. Predicting Crystal Growth by Spiral Motion. Proc. R. Soc. A 2008, in press. (95) Markov, I. V. Crystal Growth for Beginners, Fundamentals of Nucleation, Crystal Growth and Epitaxy; World Scientific: Singapore, 2003. (96) Winn, D.; Doherty, M. F. Modeling crystal shapes of organic materials grown from solution. AIChE J. 2000, 46, 1348. (97) Liu, X. Y.; Boek, E. S.; Briels, W. J.; Bennema, P. Prediction of crystal growth morphology based on structural analysis of the solid-fluid interface. Nature 1995, 374, 342. (98) Winn, D.; Doherty, M. F. Predicting the shape of organic crystals grown from polar solvents. Chem. Eng. Sci. 2002, 57, 1805. (99) Shekunov, B.; Grant, D. In Situ Optical Interferometric Studies of the Growth and Dissolution Behavior of Paracetamol (Acetaminophen). 1. Growth Kinetics. J. Phys. Chem. B 1997, 101, 3973. (100) Ristic, R.; Finnie, S.; Sheen, D.; Sherwood, J. Macro- and Micromorphology of Monoclinic Paracetamol Grown from Pure Aqueous Solution. J. Phys. Chem. B 2001, 105, 9057. (101) Vekilov, P.; Alexander, J. Dynamics of Layer Growth in Protein Crystallization. Chem. ReV. 2000, 100, 2061. (102) Voronkov, V. V. Dislocation mechanism of growth with a low kink density. SoV. Phys. Cryst. 1973, 18, 19. (103) Teng, H. H.; Dove, P. M.; Orme, C. A.; De Yoreo, J. J. Thermodynamics of Calcite Growth: Baseline for Understanding Biomineral Formation. Science 1998, 282, 724. (104) Chernov, A. A. Crystal growth science between the centuries. J. Mater. Sci.: Mater. Electron. 2001, 12, 437. (105) Chernov, A. A. Notes on interface growth kinetics 50 years after Burton, Cabrera and Frank. J. Cryst. Growth 2004, 264, 499. (106) Rashkovich, L.; Gvozdev, N.; Sil’nikova, M.; Chernov, A. Fluctuations in the step velocity and the generation of a dislocation spiral on the (101) face of monoclinic lysozyme crystals. Crystallogr. Rep. 2002, 47, 859. (107) Rashkovich, L.; Petrova, E.; Shustin, O.; Chernevich, T. Formation of a dislocation spiral on the (010) face of a potassium hydrogen phthalate crystal. Phys. Solid State 2003, 45, 400. (108) Cabrera, N.; Levine, M. M. XLV. On the dislocation theory of evaporation of crystals. Philos. Mag. 1956, 1, 450.

Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 9831 (109) Gilmer, G. H.; Bennema, P. Simulation of Crystal Growth with Surface Diffusion. J. Appl. Phys. 1972, 43, 1347. (110) Rohl, A. L. Computer prediction of crystal morphology. Curr. Opin. Solid State Mater. Sci. 2003, 7, 21. (111) van der Eerden, J. P. J. M. Molecular simulations of crystal growth processes. In Crystal GrowthsFrom Fundamentals to Technology; Mu¨ller, G., Me´tois, J. J., Rudolph, P., Eds.; Elsevier: Amsterdam, 2004; pp 187213. (112) Boerrigter, S.; Josten, G.; vandeStreek, J.; Hollander, F.; Los, J.; Cuppen, H.; Bennema, P.; Meekes, H. MONTY: Monte Carlo Crystal Growth on Any Crystal Structure in Any Crystallographic Orientation: Application to Fats. J. Phys. Chem. A 2004, 108, 5894. (113) Zepeda-Ruiz, L. A.; Maiti, A.; Gee, R.; Gilmer, G. H.; Weeks, B. L. Size and habit evolution of PETN crystals-a lattice Monte Carlo study. J. Cryst. Growth 2006, 291, 461. (114) Piana, S.; Reyhani, M.; Gale, J. D. Simulating micrometre-scale crystal growth from solution. Nature 2005, 438, 70. (115) Anderson, M. W.; Agger, J. R.; Meza, L. I.; Chong, C. B.; Cundy, C. S. Crystal growth in nanoporous framework materials. Faraday Discuss. 2007, 136, 143. (116) Rodrguez-hornedo, N.; Murphy, D. Significance of controlling crystallization mechanisms and kinetics in pharmaceutical systems. J. Pharm. Sci. 1999, 88, 651. (117) Mehta, S. C.; Bernardo, P. D.; Higuchi, W. I.; Simonelli, A. P. Rate of crystal growth of sulfathiazole and methylprednisolone. J. Pharm. Sci. 1970, 59, 638. (118) Kind, M.; Mersmann, A. On supersaturation during mass crystallization from solution. Chem. Eng. Technol. 1990, 13, 50. (119) Li, L.; Rodriguez-Hornedo, N. Growth kinetics and mechanism of glycine crystals. J. Cryst. Growth 1992, 121, 33. (120) Veesler, S.; Lafferrere, L.; Garcia, E.; Hoff, C. Phase Transitions in Supersaturated Drug Solution. Org. Process Res. DeV. 2003, 7, 983. (121) Onuma, K.; Kameyama, T.; Tsukamoto, K. In situ study of surface phenomena by real time phase shift interferometry. J. Cryst. Growth 1994, 137, 610. (122) McLoughlin, M. J.; Mays, T. J.; Price, R. Growth Spiral Activity and Step Velocities on a Crystal Surface. Phys. ReV. Lett. 2005, 95, 115504. (123) Rosenberger, F. Protein crystallization. J. Cryst. Growth 1996, 166, 40. (124) Vekilov, P. G.; Ataka, M.; Katsura, T. Growth process of protein crystals revealed by laser Michelson interferometry investigation. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1995, 51, 207. (125) Chernov, A. A. Formation of crystals in solutions. Contemp. Phys. 1989, 30, 251. (126) Rashkovich, L. N.; Shekunov, B. Y. Morphology of growing vicinal surface; prismatic faces of ADP and KDP crystals in solutions. J. Cryst. Growth 1990, 100, 133. (127) Bazzana, S.; Dumrul, S.; Warzywoda, J.; Hsiao, L.; Klass, L.; Knapp, M.; Rains, J. A.; Stein, E. M.; Sullivan, M. J.; West, C. M.; Woo, J. Y.; Sacco, A., Jr. Observations of layer growth in synthetic zeolites by field emission scanning electron microscopy. Stud. Surf. Sci. Catal. 2002, 142, 117. (128) Dumrul, S.; Bazzana, S.; Warzywoda, J.; Biederman, R. R.; Sacco, A. Imaging of crystal growth-induced fine surface features in zeolite A by atomic force microscopy. Microporous Mesoporous Mater. 2002, 54, 79. (129) Yilmaz, B.; Miraglia, P. Q.; Warzywoda, J.; Sacco, A. Synthesis of titanosilicate ETS-4 with controlled morphology and investigation of its crystallization kinetics. Microporous Mesoporous Mater. 2004, 71, 167. (130) Miraglia, P. Q.; Yilmaz, B.; Warzywoda, J.; Sacco, A., Jr. Surface growth mechanisms and structural faulting in the growth of large single and spherulitic titanosilicate ETS-4 crystals. J. Cryst. Growth 2004, 270, 674. (131) Hillner, P. E.; Gratz, A. J.; Manne, S.; Hansma, P. K. Atomicscale imaging of calcite growth and dissolution in real time. Geology 1992, 20, 359. (132) Teng, H. H.; Dove, P. M.; De Yoreo, J. J. Kinetics of calcite growth: surface processes and relationships to macroscopic rate laws. Geochim. Cosmochim. Acta 2000, 64, 2255. (133) De Yoreo, J. J.; Burnham, A. K.; Whitman, P. K. Developing KH2PO4 and KD2PO4 crystals for the world’s most powerful laser. Int. Mater. ReV. 2002, 47, 113. (134) Thomas, T. N.; Land, T. A.; Martin, T.; Casey, W. H.; DeYoreo, J. J. AFM investigation of step kinetics and hillock morphology of the 100 face of KDP. J. Cryst. Growth 2004, 260, 566. (135) Qiu, S. R.; Wierzbicki, A.; Orme, C. A.; Cody, A. M.; Hoyer, J. R.; Nancollas, G. H.; Zepeda, S.; De Yoreo, J. J. From the Cover: Molecular modulation of calcium oxalate crystallization by osteopontin and citrate. PNAS 2004, 101, 1811.

(136) Touryan, L. A.; Lochhead, M. J.; Marquardt, B. J.; Vogel, V. Sequential switch of biomineral crystal morphology using trivalent ions. Nat. Mater. 2004, 3, 239. (137) Jung, T.; Sheng, X. X.; Choi, C. K.; Kim, W. S.; Wesson, J. A.; Ward, M. D. Probing crystallization of calcium oxalate monohydrate and the role of macromolecule additives with in situ atomic force microscopy. Langmuir 2004, 20, 8587. (138) Weisenhorn, A. L.; Dougall, J. E. M.; Gould, S. A. C.; Cox, S. D.; Wise, W. S.; Massie, J.; Maivald, P.; Elings, V. B.; Stucky, G. D.; Hansma, P. K. Imaging and Manipulating Molecules on a Zeolite Surface with an Atomic Force Microscope. Science 1990, 247, 1330. (139) Agger, J.; Pervaiz, N.; Cheetham, A.; Anderson, M. Crystallization in Zeolite A Studied by Atomic Force Microscopy. J. Am. Chem. Soc. 1998, 120, 10754. (140) Anderson, M. W.; Ohsuna, T.; Sakamoto, Y.; Liu, Z.; Carlsson, A.; Terasaki, O. Modern microscopy methods for the structural study of porous materials. Chem. Commun. 2004, 907. (141) Wakihara, T.; Sugiyama, A.; Okubo, T. Crystal growth of faujasite observed by atomic force microscopy. Microporous Mesoporous Mater. 2004, 70, 7. (142) Durbin, S. D.; Carlson, W. E.; Saros, M. T. In situ studies of protein crystal growth by atomic force microscopy. J. Phys. D: Appl. Phys. 1993, 26, B128. (143) Yip, C. M.; Ward, M. D. Atomic force microscopy of insulin single crystals: direct visualization of molecules and crystal growth. Biophys. J. 1996, 71, 1071. (144) McPherson, A.; Malkin, A. J.; Kuznetsov, Y. G. Atomic force microscopy in the study of macromolecular crystal growth. Annu. ReV. Biophys. Biomol. Struct. 2000, 29, 361. (145) Thompson, C.; Davies, M. C.; Roberts, C. J.; Tendler, S. J. B.; Wilkinson, M. J. The effects of additives on the growth and morphology of paracetamol (acetaminophen) crystals. Int. J. Pharm. 2004, 280, 137. (146) Keel, T.; Thompson, C.; Davies, M.; Tendler, S.; Roberts, C. AFM studies of the crystallization and habit modification of an excipient material, adipic acid. Int. J. Pharm. 2004, 280, 185. (147) Sours, R.; Zellelow, A.; Swift, J. An in Situ Atomic Force Microscopy Study of Uric Acid Crystal Growth. J. Phys. Chem. B 2005, 109, 9989. (148) Stoica, C.; van Enckevort, W.; Meekes, H.; Vlieg, E. Interlaced spiral growth and step splitting on a steroid crystal. J. Cryst. Growth 2007, 299, 322. (149) Chernov, A. A.; Rashkovich, L. N.; Yaminski, I. V.; Gvozdev, N. V. Kink kinetics, exchange fluxes, 1D ‘nucleation’ and adsorption on the (010) face of orthorhombic lysozyme crystals. J. Phys.: Condens. Matter 1999, 11, 9969. (150) Chernov, A.; Rashkovich, L.; Vekilov, P. Steps in solution growth: dynamics of kinks, bunching and turbulence. J. Cryst. Growth 2005, 275, 1. (151) Rashkovich, L.; De Yoreo, J.; Orme, C.; Chernov, A. In situ atomic force microscopy of layer-by-layer crystal growth and key growth concepts. Crystallogr. Rep. 2006, 51, 1063. (152) Georgiou, D. K.; Vekilov, P. G. A fast response mechanism for insulin storage in crystals may involve kink generation by association of 2D clusters. PNAS 2006, 103, 1681. (153) Davis, K. J.; Dove, P. M.; De Yoreo, J. J. The Role of Mg2+ as an Impurity in Calcite Growth. Science 2000, 290, 1134. (154) Danesh, A.; Chen, X.; Davies, M.; Roberts, C.; Sanders, G.; Tendler, S.; Williams, P.; Wilkins, M. Polymorphic Discrimination Using Atomic Force Microscopy: Distinguishing between Two Polymorphs of the Drug Cimetidine. Langmuir 2000, 16, 866. (155) Wells, A. F. Crystal Habit and Internal Structure-I. Philos. Mag. 1946, 37, 184. (156) Wells, A. Abnormal and Modified Crystal Growth. Introductory Paper. Discuss. Faraday Soc. 1949, 5, 197–201. (157) Davey, R. J.; Milisavljevic, B.; Bourne, J. R. Solvent interactions at crystal surfaces: the kinetic story of R-resorcinol. J. Phys. Chem. 1988, 92, 2032. (158) Davey, R. J.; Mullin, J. W.; Whiting, M. J. L. Habit Modification of Succinic Acid Crystals Grown from Different Solvents. J. Cryst. Growth 1982, 58, 304. (159) Gordon, R. E.; Amin, S. I. Crystallization of Ibuprofen, U.S. Patent 4,476,248, 1984. (160) Davey, R. J. The role of the solvent in crystal growth from solution. J. Cryst. Growth 1986, 76, 637. (161) Lahav, M.; Leiserowitz, L. The effect of solvent on crystal growth and morphology. Chem. Eng. Sci. 2001, 56, 2245. (162) Lechuga-Ballesteros, D.; Rodriguez-Hornedo, N. The Influence of Additives on the Growth-Kinetics and Mechanism of L-Alanine Crystals. Int. J. Pharm. 1995, 115, 139.

9832 Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 (163) Davey, R. J.; Black, S. N.; Logan, D.; Maginn, S. J.; Fairbrother, J. E.; Grant, D. J. W. Structural and Kinetic Features of Crystal-Growth InhibitionsAdipic Acid Growing in the Presence of N-Alkanoic Acids. J. Chem. Soc., Faraday Trans. 1992, 88, 3461. (164) Garti, N.; Zour, H. The effect of surfactants on the crystallization and polymorphic transformation of glutamic acid. J. Cryst. Growth 1997, 172, 486. (165) Hendriksen, B. A.; Grant, D. J. W.; Meenan, P.; Green, D. A. Crystallisation of paracetamol (acetaminophen) in the presence of structurally related substances. J. Cryst. Growth 1998, 183, 629. (166) Weissbuch, I.; Addadi, L.; Berkovitch-Yellin, Z.; Gati, E.; Weinstein, S.; Lahav, M.; Leiserowitz, L. Centrosymmetric Crystals for the Direct Assignment of the Absolute-Configuration of Chiral Moleculess Application to the Alpha-Amino-Acids by Their Effect on Glycine Crystals. J. Am. Chem. Soc. 1983, 105, 6615. (167) Michaels, A. S.; Tausch, F. W. Modification of Growth Rate and Habit of Adipic Acid Crystals with Surfactants. J. Phys. Chem. 1961, 65, 1730. (168) Klug, D. L.; van Mil, J. H. Purification of Adipic Acid, U.S. Patent 5,296,639, 1994. (169) Weissbuch, I.; Addadi, L.; Lahav, M.; Leiserowitz, L. Molecular Recognition at Crystal Interfaces. Science 1991, 253, 637. (170) Michaels, A. S.; Colville, A. R. The effect of surface active agents on crystal growth rate and crystal habit. J. Phys. Chem. 1960, 64, 13. (171) DeVries, A. L. Glycoproteins as Biological Antifreeze Agents in Antarctic Fishes. Science 1971, 172, 1152. (172) Raymond, J. A.; DeVries, A. L. Adsorption Inhibition as a Mechanism of Freezing Resistance in Polar Fishes. Proc. Natl. Acad. Sci., U.S.A. 1977, 74, 2589. (173) Cabrera, N.; Vermilyea, D. The Growth of Crystals from Solution. In Growth and Perfection of Crystals; Doremus, R., Roberts, B., Turnbull, D., Eds.; John Wiley & Sons: New York, 1958. (174) Land, T.; Martin, T.; Potapenko, S.; Palmore, G.; De Yoreo, J. Recovery of surfaces from impurity poisoning during crystal growth. Nature 1999, 399, 442. (175) Chernov, A. A. The spiral growth of crystals. SoViet Phys.-Usp. 1961, 4, 116. (176) Cabrera, N.; Coleman, R. V. Theory of Crystal Growth from the Vapor. In The Art and Science of Growing Crystals; Gilman, J. J., Ed.; Wiley: New York, 1963. (177) Clydesdale, G.; Roberts, K. J.; Docherty, R. Modeling the Morphology of Molecular-Crystals in the Presence of Disruptive TailorMade Additives. J. Cryst. Growth 1994, 135, 331. (178) Clydesdale, G.; Roberts, K. J.; Lewitas, K.; Docherty, R. Modeling the Morphology of Molecular-Crystals in the Presence of Blocking TailorMade Additives. J. Cryst. Growth 1994, 141, 443. (179) Hartman, P. Structure and Morphology. In Crystal Growth: An Introduction; Hartman, P., Ed.; North-Holland Publishing Co.: Amsterdam, 1973; pp 367-402. (180) Berkovitch-Yellin, Z. Toward an ab initio derivation of crystal morphology. J. Am. Chem. Soc. 1985, 107, 8239. (181) Lighthill, M. J.; Whitham, G. B. On Kinematic Waves. 1. Flood Movement in Long Rivers. Proc. R. Soc. London, Ser. A 1955, 229, 281. (182) Lighthill, M. J.; Whitham, G. B. On Kinematic Waves. 2. A Theory of Traffic Flow on Long Crowded Roads. Proc. R. Soc. London, Ser. A 1955, 229, 317. (183) Hamaguchi, S.; Dalvie, M. Microprofile Simulations for PlasmaEtching with Surface Passivation. J. Vac. Sci. Technol., A 1994, 12, 2745. (184) Shankland, N.; Florence, A. J.; Cox, P. J.; Sheen, D. B.; Love, S. W.; Stewart, N. S.; Wilson, C. C. Crystal morphology of ibuprofen predicted from single-crystal pulsed neutron diffraction data. Chem. Commun. 1996, 855. (185) Cano, H.; Gabas, N.; Canselier, J. Experimental study on the ibuprofen crystal growth morphology in solution. J. Cryst. Growth 2001, 224, 335. (186) Randolph, A. D.; Larson, M. A. Theory of Particulate Processes: Analysis and Techniques of Continuous Crystallization; Academic Press: New York, 1988. (187) Ramkrishna, D. Population BalancessTheory and Applications to Particulate Systems in Engineering; Academic Press: San Diego, CA, 2000. (188) Iveson, S. M. Limitations of one-dimensional population balance models of wet granulation processes. Powder Technol. 2002, 124, 219. (189) Ma, D. L.; Braatz, R. D.; Tafti, D. K. Compartmental modeling of multidimensional crystallization. Int. J. Mod. Phys. B 2002, 16, 383. (190) Ma, D. L.; Tafti, D. K.; Braatz, R. D. Optimal control and simulation of multidimensional crystallization processes. Comput. Chem. Eng. 2002, 26, 1103.

(191) Puel, F.; Marchal, P.; Klein, J. Habit transient analysis in industrial crystallization using two dimensional crystal sizing technique. Chem. Eng. Res. Des. 1997, 75, 193. (192) Puel, F.; Fevotte, G.; Klein, J. P. Simulation and analysis of industrial crystallization processes through multidimensional population balance equations. Part 1: A resolution algorithm based on the method of classes. Chem. Eng. Sci. 2003, 58, 3715. (193) Puel, F.; Fevotte, G.; Klein, J. P. Simulation and analysis of industrial crystallization processes through multidimensional population balance equations. Part 2: A study of semi-batch crystallization. Chem. Eng. Sci. 2003, 58, 3729. (194) Oullion, M.; Puel, F.; Fevotte, G.; Righini, S.; Carvin, P. Industrial batch crystallization of a plate-like organic product. In situ monitoring and 2D-CSD modelling: Part 1: Experimental study. Chem. Eng. Sci. 2007, 62, 820. (195) Oullion, M.; Puel, F.; Fevotte, G.; Righini, S.; Carvin, P. Industrial batch crystallization of a plate-like organic product. In situ monitoring and 2D-CSD modelling. Part 2: Kinetic modelling and identification. Chem. Eng. Sci. 2007, 62, 833. (196) Gerstlauer, A.; Mitrovic, A.; Motz, S.; Gilles, E. D. A population model for crystallization processes using two independent particle properties. Chem. Eng. Sci. 2001, 56, 2553. (197) Briesen, H. Simulation of crystal size and shape by means of a reduced two-dimensional population balance model. Chem. Eng. Sci. 2006, 61, 104. (198) Zhang, Y.; Doherty, M. F. Simultaneous prediction of crystal shape and size for solution crystallization. AIChE J. 2004, 50, 2101. (199) Ma, C. Y.; Wang, X. Z.; Roberts, K. J. Morphological population balance for modeling crystal growth in face directions. AIChE J. 2008, 54, 209. (200) Ma, D.; Tafti, D.; Braatz, R. High-Resolution Simulation of Multidimensional Crystal Growth. Ind. Eng. Chem. Res. 2002, 41, 6217. (201) Gunawan, R.; Fusman, I.; Braatz, R. D. High resolution algorithms for multidimensional population balance equations. AIChE J. 2004, 50, 2738. (202) Pinto, M. A.; Immanuel, C. D.; Doyle, F. J., III. A feasible solution technique for higher-dimensional population balance models. Comput. Chem. Eng. 2007, 31, 1242. (203) Qamar, S.; Ashfaq, A.; Warnecke, G.; Angelov, I.; Elsner, M.; Seidel-Morgenstern, A. Adaptive high-resolution schemes for multidimensional population balances in crystallization processes. Comput. Chem. Eng. 2007, 31, 1296. (204) Chakraborty, J.; Kumar, S. A new framework for solution of multidimensional population balance equations. Chem. Eng. Sci. 2007, 62, 4112. (205) Gunawan, R.; Ma, D. L.; Braatz, M. F. R. D. Identification of kinetic parameters in multidimensional crystallization processes. Int. J. Mod. Phys. B 2002, 16, 367. (206) Ruf, A.; Worlitschek, J.; Mazzotti, M. Modeling and Experimental Analysis of PSD Measurements through FBRM. Part. Part. Syst. Charact. 2000, 17, 167. (207) Scholl, J.; Bonalumi, D.; Vicum, L.; Mazzotti, M.; Muller, M. In Situ Monitoring and Modeling of the Solvent-Mediated Polymorphic Transformation of L-Glutamic Acid. Cryst. Growth Des. 2006, 6, 881. (208) Kempkes, M.; Eggers, J.; Mazzotti, M. Measurement of particle size and shape by FBRM and in situ microscopy. Chem. Eng. Sci. 2008 634656. (209) Calderon De Anda, J.; Wang, X.; Roberts, K. Multi-scale segmentation image analysis for the in-process monitoring of particle shape with batch crystallisers. Chem. Eng. Sci. 2005, 60, 1053. (210) Calderon De Anda, J.; Wang, X. Z.; Lai, X.; Roberts, K. J.; Jennings, K. H.; Wilkinson, M. J.; Watson, D.; Roberts, D. Real-time product morphology monitoring in crystallization using imaging technique. AIChE J. 2005, 51, 1406. (211) Calderon De Anda, J.; Wang, X.; Lai, X.; Roberts, K. Classifying organic crystals via in-process image analysis and the use of monitoring charts to follow polymorphic and morphological changes. J. Process Control 2005, 15, 785. (212) Wang, X. Z.; Calderon De Anda, J.; Roberts, K. J.; Li, R. F.; Thomson, G. B.; White, G. Advances in on-line monitoring and control of the morphological and polymorphic forms of organic crystals grown from solution. Kona 2005, 23, 69. (213) Li, R. F.; Thomson, G. B.; White, G.; Wang, X. Z.; Calderon De Anda, J.; Roberts, K. J. Integration of crystal morphology modeling and on-line shape measurement. AIChE J. 2006, 52, 2297. (214) Ma, C.; Wang, X.; Roberts, K. Multi-dimensional population balance modeling of the growth of rod-like L-glutamic acid crystals using growth rates estimated from in-process imaging. AdV. Powder Technol. 2007, 18, 707.

Ind. Eng. Chem. Res., Vol. 47, No. 24, 2008 9833 (215) Wang, X. Y.; Calderon De Anda, J.; Roberts, K. J. Real-time measurement of the growth rates of individual crystal facets using imaging and image analysissA feasibility study on needle-shaped crystals of L-glutamic acid. Chem. Eng. Res. Des. 2007, 85, 921. (216) Larsen, P.; Rawlings, J.; Ferrier, N. An algorithm for analyzing noisy, in situ images of high-aspect-ratio crystals to monitor particle size distribution. Chem. Eng. Sci. 2006, 61, 5236. (217) Larsen, P.; Rawlings, J.; Ferrier, N. Model-based object recognition to measure crystal size and shape distributions from in situ video images. Chem. Eng. Sci. 2007, 62, 1430. (218) Pollanen, K.; Hakkinen, A.; Reinikainen, S. P.; Louhi-Kultanen, M.;Nystrom,L.AstudyonbatchcoolingcrystallizationofsulphathiazolesProcess monitoring using ATR-FTIR and product characterization by automated image analysis. Chem. Eng. Res. Des. 2006, 84, 47. (219) Patience, D. B.; Rawlings, J. B. Particle-shape monitoring and control in crystallization processes. AIChE J. 2001, 47, 2125. (220) Dharmayat, S.; Calderon De Anda, J.; Hammond, R. B.; Lai, X.; Roberts, K. J.; Wang, X. Z. Polymorphic transformation of L-glutamic acid monitored using combined on-line video microscopy and X-ray diffraction. J. Cryst. Growth 2006, 294, 35. (221) Bernard-Michel, B.; Pons, M. N.; Vivier, H. Quantification, by image analysis, of effect of operational conditions on size and shape of precipitated barium sulphate. Chem. Eng. J. 2002, 87, 135. (222) Pons, M.-N.; Vivier, H.; Delcour, V.; Authelin, J.-R.; PailleresHubert, L. Morphological analysis of pharmaceutical powders. Powder Technol. 2002, 128, 276. (223) Faria, N.; Pons, M. N.; Feyo de Azevedo, S.; Rocha, F. A.; Vivier, H. Quantification of the morphology of sucrose crystals by image analysis. Powder Technol. 2003, 133, 54.

(224) Zhang, D.; Lu, G. Review of shape representation and description techniques. Pattern Recogn. 2004, 37, 1. (225) Yu, Z. Q.; Chow, P. S.; Tan, R. B. Quantification of particle morphology by boundary Fourier transform and generic Fourier transform. Chem. Eng. Sci. 2007, 62, 3777. (226) Wang, X. Z.; Roberts, K. J.; Ma, C. Crystal growth measurement using 2D and 3D imaging and the perspectives for shape control. Chem. Eng. Sci. 2008, 63, 1173. (227) Yang, G.; Kubota, N.; Sha, Z.; Louhi-Kultanen, M.; Wang, J. Crystal Shape Control by Manipulating Supersaturation in Batch Cooling Crystallization. Cryst. Growth Des. 2006, 6, 2799. (228) Hermtxanto, M. W.; Chiu, M.-S.; Woo, X.-Y.; Braatz, R. D. Robust optimal control of polymorphic transformation in batch crystallization. AIChE J. 2007, 53, 2643. (229) Fujiwara, M.; Nagy, Z. K.; Chew, J. W.; Braatz, R. D. Firstprinciples and direct design approaches for the control of pharmaceutical crystallization. J. Process Control 2005, 15, 493. (230) Larsen, P.; Patience, D.; Rawlings, J. Industrial crystallization process control. Control Syst. Mag., IEEE 2006, 26, 70. (231) Yu, Z. Q.; Chew, J. W.; Tan, R. B. H. Recent advances in crystallization control. Chem. Eng. Res. Des. 2007, 85, 893.

ReceiVed for reView June 6, 2008 ReVised manuscript receiVed August 13, 2008 Accepted September 2, 2008 IE800900F

Crystal Shape Engineering - Industrial & Engineering Chemistry

Recommend Documents